Serialized snapshot blob maps corrupt arbitrary binary summary blobs

[anthony-murphy]

Describe the bug

Container-loader serialized-state paths assume structural snapshot blobs are UTF-8 text. SummaryType.Blob, however, accepts raw Uint8Array content. Arbitrary binary bytes can therefore be corrupted when a summary is converted to a snapshot, serialized through a string-valued blob map, and later rehydrated.

Affected paths include detached-container serialization/rehydration and frozen full-container capture of structural snapshot blobs. Converting invalid UTF-8 bytes to a JavaScript string replaces bytes with U+FFFD; encoding that string later cannot recover the original payload.

This surfaced while reviewing PR #27880. That feature should retain its local base64 workaround rather than expanding scope into the generic loader serialization format.

To Reproduce

Steps to reproduce the behavior:

1. Create a summary containing SummaryType.Blob with bytes such as Uint8Array([0x00, 0x7f, 0x80, 0xc3, 0x28, 0xff]).
2. Serialize it through detached-container state or capture it as a structural blob in frozen full-container state.
3. JSON round-trip and rehydrate the state.
4. Read the blob and compare it byte-for-byte with the original payload.
5. Observe replacement-byte sequences where invalid UTF-8 bytes were present.

Expected behavior

Serialized snapshot formats should preserve arbitrary blob bytes. Binary encoding should be explicit at the JSON boundary (for example, a versioned base64 encoding marker), with backward-compatible decoding for existing marker-less UTF-8 states. Paths whose contracts guarantee textual content may continue using UTF-8 explicitly.

Logs

N/A