Fix: Solana memo toBytes() corrupts hex-looking memos

[KabaDH]

Fix: Solana memo toBytes() corrupts hex-looking memos

Summary

MemoLayout.toBytes() encoded the memo with the hybrid StringUtils.toBytes()
(hex-or-UTF-8 by content sniffing), while fromBuffer and toJson treat a memo
as a plain UTF-8 string. Memos whose text happens to look like hex were silently
corrupted, and some produced invalid UTF-8 that the SPL Memo program rejects.

This PR switches the write path to a plain UTF-8 encode, making it symmetric with
the read path and consistent with the Memo program semantics.

Changes

• lib/solana/src/instructions/memo/layouts/memo.dart

   @override
   List<int> toBytes() {
-    return StringUtils.toBytes(memo);
+    return StringUtils.encode(memo); // UTF-8, symmetric with fromBuffer
   }

Impact

• Normal text memos (URLs, sentences, anything with non-hex chars or odd length)
  are unaffected — they already used the UTF-8 branch.
• Behavior changes only for memos matching ^(0x|0X)?([0-9A-Fa-f]{2})+$: the
  literal text is now preserved instead of being reinterpreted as raw bytes.
• Callers that relied on passing a hex string to embed raw bytes should pass the
  decoded bytes explicitly; that was undocumented behavior that could yield
  invalid UTF-8.

Testing

Added test/solana/tests/memo/layout_test.dart covering round-trip, byte-level
UTF-8, multi-byte characters, empty memo, and toJson consistency. All pass; the
suite fails on the pre-fix code.

[mrtnetwork]

@KabaDH Hi, I think we should change the memo variable to bytes (or a hex string) instead of UTF-8. When we use allowInvalidOrMalformed with non-UTF-8 data, the decoded value can produce different bytes, which isn't acceptable in some situations, such as Web3, where the exact byte representation must be preserved.

[KabaDH]

@mrtnetwork Done.
MemoLayout now stores raw List<int> memoBytes instead of a String, so the exact bytes are preserved. Added
fromString + memo getter for text; toJson falls back to 0x hex like ed25519.
Note: breaking — constructor now takes memoBytes. Tests cover the non-UTF-8 round-trip.