In addition to validating all decoders for memory safety we can also
validate that the vertex encoder in particular round-trips safely. Index
encoder may rotate triangles so it's not a perfect binary match, and
index sequence encoder is rarely used and very simple, but vertex
encoder is more complicated and may become more so with v1.
This code relied on cpuid; since the codecs have been extensively fuzzed
and scalar codecs are easier to verify by comparison, just disable the
cpuid fallback here as well.
All codecs must be resistant to malicious inputs; while the decoders
aren't guaranteed to provide a valid result, they must be able to decode
the input without reading past the input bounds or writing past the
output bounds.
This change adds a fuzzing driver that attempts to prove this for both
SIMD and scalar implementations of the decoders.
Arseny Kapoulkine