Added subtitle track duration to manual search to tell same-language tracks apart

[andersonviudes]

Summary

• Adds a per-track duration hint to embedded subtitle results in manual search, so tracks sharing the same language show up as distinguishable entries instead of duplicates.
• Reads the per-track MKV DURATION tag (falling back to the language-suffixed DURATION-eng tag, then the container duration) and surfaces it as Duration: h:mm:ss in the release info shown to the user.
• Small refactor of release-info assembly in manual.py into _build_release_info() to make room for the new hint; no behavior change for existing single-track-per-language cases.

Context

When an MKV has multiple subtitle tracks in the same language, embeddedsubtitles.py only exposed stream.duration, which reflects the container's total duration for every stream — so every same-language track looked identical in manual search, with nothing to tell them apart.

Per-track MKV DURATION tags actually differ between tracks and are the right signal here. For same-language tracks, mkvmerge/ffprobe often reports this tag language-suffixed instead (e.g. DURATION-eng), which the vendored fese library already exposes as stream.tags.duration_eng (mirroring how it already falls back for number_of_frames/number_of_frames_eng). The fix stays fully scoped to Bazarr's own custom_libs/subliminal_patch/providers/embeddedsubtitles.py — no changes to vendored libs/fese/*.

#Exemple:

Test plan

• Added test_get_extra_release_info_mkv_tag_duration_wins_over_container_duration and test_get_extra_release_info_mkv_language_suffixed_tag_wins_over_container_duration in tests/subliminal_patch/test_embeddedsubtitles.py, plus coverage for _build_release_info() in the new tests/bazarr/test_subtitles_manual.py.
• Ran the targeted suites locally: pytest tests/subliminal_patch/test_embeddedsubtitles.py tests/bazarr/test_subtitles_manual.py → 45 passed (2 pre-existing, unrelated failures in the same file untouched by this change).
• Also includes an unrelated no log: commit adding ruff to dev-requirements.txt + a pyproject.toml excluding libs//custom_libs//frontend//migrations/, so ruff check . doesn't lint vendored code for other contributors.