js: Use UTF-8 byte length for metadata string prefixes

[ara-stock]

Problem

updateTokenMetadataInstruction in clients/js writes each borsh string length prefix using the JavaScript string length (UTF-16 code units) but then appends the UTF-8 bytes from TextEncoder. For any name, symbol or URI containing non-ASCII characters (e.g. é, ☀, emoji), the prefix is shorter than the actual byte length, so the program misreads the instruction data. With the name Café Staking ☀ the program reads a garbage length for the next field and aborts:

Error: memory allocation failed, out of memory
Program SVSPxpvHdN29nkVg9rPapPNDddN5DipNLRUFhyjFThE failed: SBF program panicked

This affects both @solana/spl-single-pool and @solana/spl-single-pool-classic, since the legacy client calls the same builder. The Rust update_token_metadata builder uses borsh and is not affected.

Summary of Changes

• Encode name, symbol and URI to UTF-8 first and use the encoded byte lengths as the length prefixes. Output is unchanged for ASCII-only strings.
• Add a js-legacy test that updates metadata with a non-ASCII name and symbol and checks both end up in the metadata account. It fails on main with the panic above and passes with this change.

I used an AI assistant (Claude) while preparing this. I reproduced the issue and checked the fix and tests myself.

🤖 Generated with Claude Code

[joncinque]

Thanks for finding and fixing this!