Fix merge validation for blank translated chunks

[cursor]

Bug and impact

• A translated chunk containing only whitespace was treated as present/non-empty during merge validation.
• merge_markdown_files() then stripped that chunk to empty content and skipped it, allowing output.md and final book artifacts to be generated with missing pages/chapters.
• Severely truncated outputs were also only warnings, so a bad translation could still ship as a partial book.

Root cause

• Validation used file byte size only and did not verify substantive text before merge.
• The merge loop ignored empty content after .strip() instead of failing.
• run_state could record blank/truncated outputs as usable state.

Fix

• Reject blank/unreadable outputs and severely truncated outputs in manifest validation.
• Fail manifest and legacy merges if a chunk trims to empty or cannot be read.
• Mark blank/truncated outputs for re-translation and prevent recording them in run_state.
• Added regression tests for manifest merge, legacy merge, and selective run-state planning/recording.

Validation

• Reproduced the pre-fix bug: whitespace-only output_chunk0002.md produced merge_ok=True and an output.md missing chunk 2.
• Verified post-fix reproduction now fails with ERROR: Blank output and does not create output.md.
• python3 -m unittest tests.test_merge_and_build.MergeMarkdownValidationTests tests.test_run_state.RunStateTests -v
• python3 -m unittest discover -s tests -p 'test_*.py' -v (210 tests, 3 skipped)
• python3 -m compileall scripts tests
• Full baseline smoke: converted standard-alice.epub, copied chunks as mock translations, merged and built HTML/DOCX/PDF successfully. EPUB still fails due to the documented Ubuntu Calibre 7.6.0 packaging bug.