rchd: say how long lent bytes have to live

rchd_feed_borrow said lent bytes had to outlive every rchd_read_step()
that returned RCHD_PENDING. The decoder is done with them sooner: once
it has decoded the hunk, which is before it asks for the next range.
Spell that out, so a caller without a mapping can lend one staging
buffer per request and refill it for the next instead of paying for
rchd_feed_at()'s copy, and say that lent bytes are used once.

Documentation only; the decoder already behaves this way.

core/libretro-common/include/formats/rchd.h

@@ -572,8 +572,12 @@ int rchd_feed_at(rchd_t *chd, uint64_t offset, int source,
  * hunk, and at the maximum hunk size that is half a megabyte of
  * memcpy the caller has already paid for once.
  *
- * @data must stay valid and unmoved until the next rchd_read_step()
- * returns something other than RCHD_PENDING. That rules out any buffer
+ * @data must stay valid and unmoved until the decoder is done with
+ * it: until a later rchd_read_step() either finishes the read (returns
+ * something other than RCHD_PENDING) or asks for a different range. A
+ * caller may therefore lend one staging buffer per request and refill
+ * it for the next one. Lent bytes are used once; asking for the same
+ * range again means supplying it again. That rules out any buffer
  * whose pages can be released underneath the decoder while it works,
  * and it rules out a base pointer that can move: a growing arena may
  * reallocate, so re-read the base after growing it and borrow from the