fix(convert): reject stale source caches

[cursor]

Bug and impact

convert.py reused {book_name}_temp caches solely by source basename. Running conversion on two different files named the same, or replacing a source file in place, could silently reuse stale input.md/chunks/outputs and build the wrong book while regenerating a clean-looking manifest/config.

Root cause

Existing-cache skips for input.html, input.md, and chunk*.md had no source identity/content check.

Fix

• Record source_fingerprint.json with source realpath, size, and SHA-256 on fresh conversion.
• Reject cached conversion/translation artifacts when the fingerprint is missing or mismatched.
• Add regression tests for valid reuse, legacy cache rejection, same-basename different source, and in-place source replacement.

Validation

• python3 -m unittest tests.test_convert.SourceCacheFingerprintTests -v
• python3 -m unittest discover -s tests -p 'test_*.py' -v (210 tests passed, 3 skipped)
• python3 -m compileall scripts tests
• Standard Alice smoke test: convert generated 38 chunks and merge/build produced HTML/DOCX/PDF; EPUB failed due the known Ubuntu Calibre 7.6.0 packaging issue documented in AGENTS.md.
• Manual CLI repro: same-basename different source cache now exits 1 with different source before conversion.