fix(build): use boolean.h, not stdbool.h, in the new headers (#258)

CI's "Check for stdbool.h usage" job scans every src/**/*.h for a direct
<stdbool.h> include and rejects it — MSVC 2005/2010 has no such header.
memtrack.h and nvmbios.h both had one; switch to <boolean.h> like the
rest of src/.

My mistake was linting only the files I touched (c89-lint.sh with
explicit paths) rather than running it bare the way CI does.

Also adds docs/memory-track.md, which nvmbios.h already referenced:
the flash address map and unlock protocol, the $2400/'_NVM' + $2404
dispatcher contract, the on-cart filesystem layout, the module quirks
worth preserving and why, and where each piece lives in this core —
with credit to skunk_mtrk (original Atari NVM BIOS v1.01 source, CC0)
and the MiSTer Jaguar core.

Full `bash scripts/c89-lint.sh` (no args) clean; stdbool scan clean;
make TEST_EXPORTS=1 test exit 0.

docs/memory-track.md

@@ -0,0 +1,134 @@
+# Memory Track (Jaguar CD save cartridge)
+
+The Memory Track is a 128 KB flash cartridge, sold separately in 1995, that
+plugs into the Jaguar's cartridge slot **while the CD unit sits on top** — the
+CD drive has its own pass-through slot. CD games use it to save settings,
+progress and high scores.
+
+Neither the Jaguar Technical Reference Manual nor the official SDK documents
+it. Everything below was reconstructed from two projects, with thanks:
+
+- **[cubanismo/skunk_mtrk](https://github.com/cubanismo/skunk_mtrk)** —
+  preserves the **original Atari NVM BIOS v1.01 68K source** (`asmnvm.s`) and
+  its February 1995 specification, *"Non Volatile Memory - Bios calls"*.
+  Released by James Jones under CC0-1.0.
+- **[MiSTer-devel/Jaguar_MiSTer](https://github.com/MiSTer-devel/Jaguar_MiSTer)** —
+  `Jaguar.sv` carries the flash detection protocol and address map in
+  comments. The only known-working open-source hardware implementation.
+
+BizHawk's Jaguar core is a Virtual Jaguar port and shares our original gaps.
+
+## Two layers
+
+Games do **not** talk to the flash chip. They call a RAM-resident BIOS module,
+which talks to the flash. Both layers have to work.
+
+### Layer 1 — the flash device
+
+| what | where |
+|---|---|
+| NVRAM data | **`$900000`–`$91FFFF`** (128 KB, plain 16-bit window) |
+| Unlock addr 1 | `$815554` (`$800000 + 4*$5555`) |
+| Unlock addr 2 | `$80AAA8` (`$800000 + 4*$2AAA`) |
+| Manufacturer ID | `$800000` → `$1F` (ATMEL) or `$01` (AMD) |
+| Device ID | `$800004` → `$D5` (AT29C010) or `$20` (AM29F010) |
+
+Unlock sequence: `$AA` → `$815554`, `$55` → `$80AAA8`, then a command to
+`$815554` — `$90` enters ID mode ("override memtrack flash in place of cart
+rom"), `$F0` undoes it, `$A0` enables writes. After the unlock write,
+`$80AAA8` reads back `$0055` (the "ROMULATOR" probe).
+
+Two things that are easy to get wrong, and that we had wrong:
+
+- **The save data is not in the `$8xxxxx` ROM window.** It is at `$900000`.
+  Mapping it over `$8xxxxx` collides with the CD BIOS.
+- **There is no MEMCON1/ROMWIDTH gating.** A plain cartridge sits at ROMWIDTH 2
+  as its *normal* width, and CD content runs at ROMWIDTH 0 — so a ROMWIDTH test
+  makes the device unreachable for discs. MiSTer gates purely on presence. The
+  device must instead claim only its own addresses so everything else still
+  reads the cartridge ROM or CD BIOS (`MTClaimsRead`/`MTClaimsWrite` in
+  `src/core/memtrack.c`).
+
+### Layer 2 — the NVM BIOS module
+
+The CD BIOS boot copies a 2 KB module from the cartridge into RAM:
+
+- **`$2400`** — magic cookie `'_NVM'` (`$5F4E564D`). This is how games detect
+  the cartridge. Vid Grid's check is at `$012B72`: `cmpi.l #'_NVM',$2400.w`.
+- **`$2404`** — dispatcher. `JSR` with an opcode and args on the stack.
+
+Opcodes: 0 Initialize, 1 Create, 2 Open, 3 Close, 4 Delete, 5 Read, 6 Write,
+7 SearchFirst, 8 SearchNext, 9 Seek, 10 Inquire. All return a 32-bit value in
+`D0`; negative means error (`ENOINIT` −1, `ENOSPC` −2, `EFILNF` −3, `EINVFN`
+−4, `ERANGE` −5, `ENFILES` −6, `EIHNDL` −7).
+
+`Initialize` must be called first and takes the application name plus a 16 KB
+scratch buffer the module may trash.
+
+#### On-cart filesystem
+
+- 512-byte blocks, 256 total. Blocks 0–7 (first 4 KB) are system.
+- Block 0 is the FAT — one byte per block: `0` free, `1` end-of-chain, `2`
+  empty file, otherwise the next block.
+- Blocks 1–7 are the directory: 199 entries × 18 bytes —
+  `startblock`(1) `numblocks`(1) `appname`(10) `filename`(6).
+- Names pack 3 characters per 16-bit word from a 40-character set
+  (`A-Z`, `0-9`, `:`, `'`, `.`, space); app names ≤ 15 chars, file names ≤ 9.
+  Anything outside the set packs as space.
+- Byte 0–1 of block 0 is a 16-bit checksum of bytes 2..511; byte 3 is the
+  used-block count.
+- A file is identified by **app name + file name**, so two games can use the
+  same file name without colliding.
+
+#### Quirks worth preserving
+
+Games were written against the shipped module, so `src/core/nvmbios.c` keeps
+its behaviour rather than an idealised version:
+
+- **A bad checksum silently reformats** the FAT and directory and reports
+  success — which is how a blank cartridge self-formats on first use.
+- **Delete always reports success.** The module recalculates the checksum
+  afterwards, clobbering `D0`, so even a missing file returns non-negative.
+- **Close does no handle validation.**
+- Only **3 file handles** exist; a 4th `Open` returns `ENFILES`.
+- `file_offset` reports `ERANGE` for offset 0 on a zero-length file (an
+  acknowledged bug in the original).
+
+## In this core
+
+| piece | file |
+|---|---|
+| Flash device + address claims | `src/core/memtrack.c` |
+| NVM BIOS module (HLE) | `src/core/nvmbios.c` |
+| Presence flag, dispatcher hook | `src/core/jaguar.c` |
+| Option, install, save wiring | `libretro.c` |
+
+Presence is CD content (or an explicitly loaded MT cart dump). The core option
+`virtualjaguar_memory_track` (default enabled) lets you emulate a console
+without the cartridge — games then warn that information cannot be saved,
+which is correct hardware behaviour.
+
+The dispatcher is hooked in the same pre-instruction callback the CD HLE jump
+table uses, and runs in **both** boot modes: the module is RAM-resident on
+hardware regardless of which CD BIOS booted the disc.
+
+Save data lives in the CD save buffer alongside the cart and CD EEPROM banks,
+so it round-trips through the frontend's `.srm`.
+
+## Tests
+
+- `test/test_memtrack.c` — flash device: address claims (including that the
+  CD BIOS at `$800000` is *not* swallowed), ID reporting, write-enable gating,
+  full 128 KB addressability without aliasing.
+- `test/test_nvmbios.c` — filesystem: round trips, app-scoped naming, handle
+  exhaustion, search, inquire accounting, every error code, and each quirk
+  above.
+
+Both run in `make test` and need no ROMs.
+
+## Status
+
+Vid Grid (USA) (Rev 1) is the first title confirmed to use it — it boots into
+its level-select with a working Save option in both boot modes. Other CD
+titles have not been individually surveyed; a title that never calls the
+module simply never sees a cartridge, which is also what happens on hardware.

src/core/memtrack.h

@@ -7,7 +7,7 @@
 
 #include <stdint.h>
 
-#include <stdbool.h>
+#include <boolean.h>
 
 #define MT_MEM_SIZE 0x20000
 

src/core/nvmbios.h

@@ -22,7 +22,7 @@
 
 #include <stdint.h>
 #include <stddef.h>
-#include <stdbool.h>
+#include <boolean.h>
 
 #define NVM_COOKIE_ADDR    0x2400
 #define NVM_DISPATCH_ADDR  0x2404