Add mock test coverage for LLMTranslator (make_chunks + translate)

[MaleicAcid]

Add tests/test_llm_translator.py with comprehensive unit and mock tests for LLMTranslator, preparing for the issue #59 pipeline decoupling refactor.

New tests

TestMakeChunks — 7 pure unit tests (no mocks)

• test_basic — standard multi-chunk split
• test_exact_fit — texts exactly fill one chunk
• test_merge_small_tail — small trailing chunk merged into previous
• test_no_merge_large_tail — large trailing chunk kept separate
• test_empty — empty input
• test_single_item — single text
• test_line_numbers — line numbers start at 1 and increment continuously across chunks

TestLLMTranslatorTranslate — 6 mock tests (no API calls)

• test_single_chunk — texts fitting in one chunk, translate_chunk called once
• test_multiple_chunks — texts spanning 2 chunks, translate_chunk called twice
• test_context_passing_between_chunks — summary from chunk N appears in chunk N+1's context
• test_length_mismatch_triggers_atomic — wrong-length translation triggers atomic_translate fallback
• test_retry_agent_used_on_primary_failure — primary agent failure activates retry agent
• test_resume_from_compare_file — translation resumes from saved intermediate JSON, skipping completed chunks

All tests mock ChunkedTranslatorAgent and ContextReviewerAgent to avoid real API calls.

[MaleicAcid]

There will be at least 2-3 additional PRs coming to refine the translation tests. Solid test coverage is a prerequisite for us to proceed with the refactoring safely and smoothly.
The world is changeing. We need move fast. @zh-plus Feel free to bounce any ideas off me whenever you want.

[zh-plus]

thanks for your pr!