chore: release 2.1.1

[bonega]

Perf and code-size release — no API changes.

Changes since v2.1.0

• ~12–15% faster bulk decode of mostly-ASCII input — decode-table entries are now niche-packed NonZeroU32 (Pack decode-table entries as niche-optimized NonZeroU32 #36).
• ~40% / ~34% smaller monomorphized bulk decode (complete / incomplete) from unaligned word reads, dropping the align_to prefix loop and small-input fallback; decode throughput unchanged (refactor: read decode words unaligned, dropping the align_to prefix #39).

Verification

• cargo test green on default, --all-features, and --no-default-features --features alloc; --no-default-features builds.
• Decode benchmarks confirmed neutral via pinned, interleaved A/B (taskset + ASLR off).

Bumps version to 2.1.1 in Cargo.toml and all README references; adds the [2.1.1] CHANGELOG section.