Changelog, API docs, enum type inference and Bits<N> width fixes

[rudib]

Summary

• Add a changelog. Adds short summaries of every published release and a detailed entry for the unreleased 0.12.0. The entry covers the wire-format fixes from Fix packing of unaligned integers, narrow integer widths and exclusive byte ranges #119 (exclusive byte ranges, wide unaligned fields) and includes an upgrade checklist.
• Document the public API. Every public item in packed_struct and packed_struct_codegen now has docs, and both crates warn on missing docs. Docs that no longer matched the code are corrected, and the README gets the same fixes. The stale docs were:
  • the LsbInteger example
  • the description of sign extension
  • the byte_types_* features
  • the attribute tables
• Fix the inferred primitive type of enums. #[derive(PrimitiveEnum)] could pick a type too small for the discriminants. For example, -3000000000 became i32, and -1 with 200 became i8, so the derive failed to compile. It now picks the smallest integer that holds both the smallest and the largest discriminant. An enum that no integer type can hold is reported as a compile error.
• Generate Bits<N> up to the full byte width. The build script used an exclusive range, so the widest type was one bit short. As a result, ReservedZero<Bits<256>> and the 64- and 256-byte equivalents failed to compile.
• Document lsb0 bit numbering (Question on LSB0 endianness #92). In lsb0, bit 0 is the least significant bit of the last byte, and endian only orders the bytes inside each field. That means bit_numbering="lsb0", endian="lsb" does not describe a little-endian register, which is what Question on LSB0 endianness #92 expected. This is a documentation gap, not a soundness bug. The docs now explain the numbering and include a runnable example for little-endian registers. New tests cover that example and pin the layout of the struct from the issue.

Test plan

• cargo test --workspace, including the new doctest and tests/issue_92.rs

🤖 Generated with Claude Code