fix: harden AudioLoader against SIGSEGV from corrupt/malformed audio frames

[xiaden]

This PR Adds defensive guards in both decodePacket() and flushPacket() to prevent segmentation faults when processing corrupted or malformed audio files:

• Check for null frame data pointers before memcpy/swr_convert
• Validate channel count matches expected layout (prevents swr misuse)
• Sanity-check nb_samples to prevent integer overflow in buffer size calc
• Skip frames that would overflow FFMPEG_BUFFER_SIZE instead of clamping
• Remove redundant std::min() clamping that masked the overflow silently

All guards emit E_WARNING with diagnostic details (codec name, sample count, channel count) and skip the offending frame rather than crashing.

Behavior

• Valid audio: No observable change (A/B tested in bulk ML tagging workload)
• Partially corrupted audio: Frames may now be skipped instead of processed
• Fully corrupted frames that previously triggered SIGSEGV: Now handled safely

Notes

• This implementation opts to skip invalid frames rather than abort processing. If a different failure strategy is preferred, I’m happy to adapt.

• The changes were implemented with AI assistance, and I’m not deeply familiar with all invariants in this codebase, so I’d appreciate careful review, particularly around buffer handling and swr assumptions.