Expose libswresample options in AudioResampler

[WyattBlue]

Closes #2262

Summary

Adds an options parameter to av.AudioResampler so callers can pass libswresample options through to the underlying resampler:

resampler = av.AudioResampler(
    format="fltp",
    layout="mono",
    rate=16000,
    options={"resampler": "soxr", "precision": "28", "cutoff": "0.95"},
)

How it works

AudioResampler builds a small filter graph (abuffer → aformat → abuffersink). The aformat filter is metadata-only — the actual resampling is done by an aresample filter, which owns the SwrContext and exposes every swr option as a child AVOption.

When options is supplied, an explicit aresample filter is inserted into the chain:

abuffer → aresample(**options) → aformat → abuffersink

instead of relying on the one FFmpeg auto-inserts. Options flow through avfilter_init_dict, which searches child contexts, so they land on the SwrContext. No new C/Cython bindings are required — this stays within the existing filter API.

Notes

• An unknown option surfaces as a clear ValueError: unused config: <name> rather than being silently ignored.
• resampler=soxr additionally requires FFmpeg to be built --enable-libsoxr; on builds without it, FFmpeg raises Invalid argument (a build-time decision, unrelated to this API). The native swr knobs (filter_size, phase_shift, cutoff, etc.) work on any build.

Tests

• test_swr_options — verifies options pass through and output params are correct.
• test_swr_options_invalid — verifies an unknown option raises a clear error.

Existing resampler tests and mypy are green.