fix: RFC 5987 encoding for Unicode filenames in Content-Disposition header

[beveradb]

Summary

• Fixes download failures for songs with non-ASCII characters (Vietnamese, Japanese, Korean, Chinese, etc.)
• The audio-separator API now properly encodes Unicode filenames in HTTP Content-Disposition headers using RFC 5987

Problem

When downloading separated audio files with Unicode filenames like Việt Nhân - Sầu Thương_(Vocals).flac, the download endpoint failed with:

'latin-1' codec can't encode character '\u1ec7' in position 23: ordinal not in range(256)

This affected any song with non-ASCII characters in artist name or title.

Solution

Updated the /download/{task_id}/{file_hash} endpoint to use RFC 5987 encoding:

# ASCII fallback for older clients
ascii_filename = "".join(c if ord(c) < 128 else "_" for c in actual_filename)
# UTF-8 encoded filename for modern clients
encoded_filename = quote(actual_filename, safe="")
content_disposition = f'attachment; filename="{ascii_filename}"; filename*=UTF-8\'\'{encoded_filename}'

Test plan

• All 204 unit/contract tests pass
• Verified encoding produces HTTP-compliant headers (can be encoded as latin-1)
• Verified Vietnamese filename "Việt Nhân - Sầu Thương" encodes correctly
• Deploy to Modal and test with real Unicode filename

---

@coderabbitai ignore

🤖 Generated with Claude Code