feat(cli): group readable subtitle cues

[LauraGPT]

Summary

• group short and continuation sentence timestamps into bounded, readable subtitle cues
• keep speaker changes, gaps over 500 ms, cues over 8 seconds, and cues over 42 characters as hard boundaries
• make readable grouping the default for CLI SRT and the SRT/VTT example, with sentence modes preserving raw model boundaries
• leave recognized text, punctuation, JSON, and TSV output unchanged

Validation

• python -m pytest -q tests/test_cli.py tests/test_generate_subtitle.py (11 passed)
• python -m compileall -q funasr/cli.py examples/subtitle/generate_subtitle.py
• git diff --check
• reproduced on the 309.6-second MP3 from srt断句功能智能化程度不高, 存在较大比例的明显不合理断句 #3539 with SenseVoiceSmall, FSMN-VAD, CT-Punc, CUDA 12.8, and an H100
• subtitle cues: 174 -> 76; one-to-two-character fragments: 3 -> 0
• pure post-processing of the same sentence result preserved text exactly and retained the first/last boundaries; maximum grouped cue was 40 characters and 5.88 seconds
• the reported 呼吸机、 / 除颤仪， / 属于二类医疗器械。 fragments become one cue without rewriting punctuation

Closes #3539