Reuse precomputed source analysis and cache FFmpeg availability to speed up optimization

[MLyte]

Motivation

• The UI flow commonly runs analyze then optimize, but optimize() always re-ran full source analysis, causing redundant work and slower time-to-results.
• Checking FFmpeg availability was repeated on each analysis call, adding an avoidable fixed startup cost.

Description

• Added an _ffmpeg_checked flag and _ensure_ffmpeg_available() in EngineService to run assert_ffmpeg_available() only once per service instance.
• Extended EngineService.optimize() with an optional source_analysis parameter and logic to reuse a provided SourceAnalysis when it matches the input file, avoiding a second analysis.
• Updated the GUI WorkerRequest and EngineWorker to carry source_analysis and wired MainWindow to pass a cached analysis when the selected input file matches the current analysis, and to invalidate the cached analysis when the user changes file.
• Added tests/test_service.py with unit tests that verify FFmpeg availability is checked once per instance and that optimize() reuses precomputed analysis without re-invoking analyze_source().

Testing

• Ran pytest -q (default) which initially failed collection due to optimaster import resolution in this environment.
• Ran PYTHONPATH=src pytest -q which executed the test suite and produced 6 passed.

---

Codex Task