fix(s3): fix binary data corruption on upload/download

[zaru]

Summary

• PutObject used c.req.text() which corrupted binary files (docx, xlsx, etc.) via lossy UTF-8 conversion
• Switch to arrayBuffer() and store body as base64, decode back to Buffer on GET using c.body()
• Fix ETag and content_length to reflect actual binary byte counts

Reproduce

# Upload Word file
curl -X PUT http://localhost:4000/s3/emulate-default/sample.docx \
  -H "Authorization: Bearer token" \
  -H "Content-Type: application/vnd.openxmlformats-officedocument.wordprocessingml.document" \
  -H "x-amz-meta-author: test" \
  --data-binary @"./sample.docx"

# Download Word file
curl http://localhost:4000/s3/emulate-default/sample.docx \
  -H "Authorization: Bearer token" -O

# Open Word file, but it's broken
open sample.docx

Changes

• c.req.text() → c.req.arrayBuffer() + base64 encoding for storage
• c.text() → c.body() + base64 decoding for retrieval
• Widen md5() type signature to accept string | Buffer
• Add binary data round-trip test

I've confirmed that base64 encoding resolves the binary corruption issue, but I'm not sure if this is the best approach for the project. If there's a more suitable way to handle binary data storage, I'd appreciate any feedback or suggestions. Happy to revise the implementation accordingly.

[vercel]

@zaru is attempting to deploy a commit to the Vercel Labs Team on Vercel.

A member of the Team first needs to authorize it.