Support more than 2 channels (surround) for Opus

[tphakala]

The Opus writer and reader currently cap audio at mono/stereo. Opus carries surround via dOps ChannelMappingFamily > 0 with a channel mapping table (stream count, coupled count, and the per-channel mapping), which AppendDops/ParseDops do not yet emit or parse (family 0 only). Multichannel Opus needs the mapping-table bytes in dOps, relaxing the channels 1..2 guard in opusm4a (and maxOpusChannels), and channel-count plumbing through the bridge. It is also gated upstream on a go-opus multistream encode/decode API.

Niche for the current audio-clip use case; filed to track.

Scope

Originally filed for Opus and FLAC. The FLAC half shipped separately (FLAC up to 8 channels, #38), so this is now Opus-only.

AAC-LC surround (channel configurations 3-7, i.e. 3.0 through 7.1) is technically valid in the container as well, but it is blocked upstream on go-aac, which is AAC-LC mono/stereo only: it cannot encode or decode more than 2 channels, and the reader only trusts the ASC channel configuration for 1..2. If go-aac ever gains multichannel support, AAC surround belongs in its own issue rather than here, since the work is entirely upstream.

[tphakala]

One guard to add to the list. #23 introduced a fourth channels 1..2 check, on the decode side: opusm4a.openStream rejects info.Channels < 1 || info.Channels > maxOpusChannels with a wrapped m4a.ErrUnsupported before building the decoder.

It matters more than its size suggests, because it is the one that gates reading a surround file rather than writing one. The other three guards are all on encode paths, so today a multichannel Opus file produced elsewhere is rejected at the door even though the container layer would demux it fine. It also sizes the decoder's sample buffer (make([]int16, maxSamplesPerChannel*channels)), so relaxing it is not purely a validation change.

[tphakala]

The FLAC half of this landed in #38: the writer and reader now support up to 8 channels (the FLAC format maximum), with the channel count carried through the sample entry and taken authoritatively from STREAMINFO on read. 5.1 and 7.1 round-trips are covered by tests.

The Opus surround half stays open and is blocked upstream: the pinned go-opus release caps both its encoder and decoder at 1-2 channels and exposes no multistream API, so emitting a dOps channel-mapping-family greater than 0 could not round-trip (nothing could encode or decode the audio). This needs a multistream encoder/decoder in go-opus first; once that exists the remaining work here is the dOps mapping-table emit/parse in AppendDops/ParseDops plus relaxing the Opus channel guards.

[tphakala]

Trimmed scope to Opus only. The FLAC half shipped in #38 (FLAC up to 8 channels), and AAC-LC surround is gated upstream on go-aac (mono/stereo only, cannot encode or decode more than 2 channels), so this issue now tracks just the Opus dOps mapping-table work. If go-aac gains multichannel support, AAC surround should be its own issue.