[NathanFlurry]

• Vendor the BARE schema parser from bare_proc 0.2.0 through a pinned regenerate script and checked-in patch, with MIT attribution
• Support explicit union tags and enum values, and report schema errors instead of panicking
• Validate schemas against the BARE specification's invariants
• Add a zero-copy runtime: verify-once views, Bytes/Str-backed owned values, list and map views, Encode/Sink, and a Versioned trait
• Replace the serde/serde_bare generator with a from-scratch generator emitting views, borrowed refs, and owned types
• Follow the BARE spec where the previous generator diverged: explicit enum values, data[N] and list<T>[N] longer than 32
• Decode strictly: reject invalid bools and optional tags, non-minimal varints, duplicate map keys, and trailing bytes
• Add a conformance suite checking byte identity against bare_proc + serde_bare and bare-ts, the spec's example vectors, rejection cases, allocation counts, differential fuzzing, and benchmarks
• Bump the Rust crates to 0.2.0 (breaking API change; wire format unchanged except for the spec fixes above)
• Decode owned values in a single pass over the input
• Add opt-in serde derives on generated owned types
• Add Versioned slice helpers under the 0.1 names, void message encoding, and envelopes whose latest value is not a schema type
• Seal versioned messages into caller-provided buffers
• Detach decoded strings and messages from their source buffer so small fields do not keep large inputs alive
• Compile the example schemas in compiler tests and drop the indoc dependency