Preserve bytes subclasses when encoding headers

[AryanRG]

Encoder.encode() silently changes header names and values supplied as ordinary bytes subclasses. For example, a subclass containing b'custom-key' is encoded as the literal text b"b'custom-key'". A subclass dictionary key containing b':method' also stops being recognized as a pseudo-header.

Use isinstance(value, bytes) in _to_bytes() to preserve the original bytes. This restores the subclass behavior that preceded the performance optimization in #276. Remove the two type ignores made unnecessary by the resulting type narrowing.

Seven regression cases cover empty and non-UTF-8 bytes, exact wire encoding and decoding in both Huffman modes, and dictionary pseudo-header ordering. All seven fail on the original source. The independent reviewer also reproduced all seven failures using the original conversion function.

Validation on macOS:

• Upstream baseline: 502 tests passed, 100% statement and branch coverage on Python 3.13.12.
• Candidate: 509 tests passed with 100% statement and branch coverage on Python 3.11.16, 3.12.15, 3.13.12 and 3.14.4.
• Ruff, strict mypy, git diff --check, check-manifest, wheel/sdist builds and twine check passed.
• HTML documentation build succeeded with the existing unresolved our github page target at docs/source/installation.rst:10. Existing Hypothesis initialization warnings remain.
• Separate automated AI review: no blockers, with 8 focused tests, Ruff, mypy and whitespace checks independently passing.

AI disclosure: the implementation, regression tests and this description were developed by AI agents. A separate AI agent performed an automated review. No human review is claimed; this PR is a draft pending maintainer review. Hosted CI is not yet verified.