Add fixed-width primitive codecs (sized ints, floats, fixed)

[bavulapati]

Adds the fixed-width primitive codecs, matching the JS compact-encoding wire format byte-for-byte:

• Sized unsigned ints: uint8, uint16, uint24, uint40, uint48, uint56 (N-byte little-endian).
• Sized signed ints: int24, int40, int48, int56 (zigzag over the fixed-width uint).
• Floats: float32, float64 (IEEE-754 little-endian via struct).
• Fixed byte blocks: fixed(n) factory plus fixed32 (32 bytes) and fixed64 (64 bytes), no length prefix.

Validation reuses the existing _validate_uint (rejects values above MAX_SAFE_INTEGER, matching JS) plus a per-type width bound; out-of-range encodes raise ValueError, truncated decodes raise OutOfBounds. Each family is a single parameterized codec class. Every codec has byte-exact unit tests (cross-checked against the JS reference) and a round-trip row.

No new dependencies (struct is stdlib); no existing codec is changed.