Support size limits in buffered-file readline

[Ra1nhy]

Summary

Support the optional size argument on AbstractBufferedFile.readline, following the IOBase contract.

The current signature rejects every positional size with TypeError. This also breaks an existing standard-library consumer: http.client.parse_headers() calls fp.readline(_MAXLINE + 1). A regression test exercises that parser against an AbstractBufferedFile backed by MemoryFileSystem, without network access.

Positive sizes now bound the returned bytes, stopping at a newline or EOF and preserving any unread bytes after the newline. No-argument, negative and None calls keep the existing readuntil path. Zero-sized calls retain the existing closed-file checks. Invalid non-integer sizes are rejected before consuming data.

Tests compare returned bytes and file positions with io.BytesIO across three cache types, empty content, line boundaries, repeated calls and seeking. A changelog entry is included. This does not change readlines(hint) or backend-specific overrides.

Validation

CPython 3.12.14, macOS, against upstream 685f62e:

• New regression/control cases on unchanged production: 79 failed, 5 passed. With the fix: 84 passed.
• Complete test_spec.py: 466 passed, 66 skipped, 1 xfailed (original baseline: 382 passed, 66 skipped, 1 xfailed).
• Specification, core, caches, memory and local modules together: 1,097 passed, 91 skipped, 1 xfailed; one existing fork-related deprecation warning.
• Ruff 0.14.3 check and format check passed for both Python files.
• Strict Sphinx build was attempted but failed on an unchanged changelog inline-literal warning. A separate untouched upstream export reproduced the same warning at line 15 (line 18 with this entry).

Full optional/cloud/Docker/downstream tests and other Python/OS environments were not run. This is a reproducible API compatibility fix, not a claim of a production incident or measured performance improvement.