feat(solana): add SPL Memo program visualizer preset

[kyle-anchorage]

1. Why this PR exists

The SPL Memo program (MemoSq4gqABAXKb96qnH8TysNcWxMyWCqXgDLGmfcHr) is one of the most common instructions in Solana transactions, but the parser had no visualizer for it. A memo instruction fell through to the unknown_program catch-all and rendered as raw hex, so a signer could not read the memo text they were about to sign. The program was already a trusted, canonically-named entry (Memo Program) in builtin_programs.rs, but nothing drove its instruction-body rendering.

No ticket.

2. What changed

• Added a memo preset (MemoVisualizer) that decodes a memo instruction's data as its UTF-8 text and renders it under VisualizerKind::Payments.
• Memo is a native, non-Anchor program: its instruction data carries no 8-byte discriminator and no borsh args — the entire data buffer is the memo. So this preset follows the native-program pattern of compute_budget / associated_token_account (direct decode), not the Anchor IDL/discriminator path used by dflow_aggregator and the dApp presets. No JSON IDL file is bundled.
• Rendering: title Memo; condensed view shows Program: Memo Program plus the memo text; expanded view adds the resolved Program ID and the raw-data hex.
• Decoding is total: empty data renders (empty memo), non-UTF-8 data renders (non-UTF-8 data; see Raw Data), and the raw bytes stay visible in the Raw Data field in every case.
• Registered only the v2 program ID (the active one). Memo v1 (Memo1Uhk...) has an identical interface and could be added with a one-line config change.

3. Why this is safe

Backward compatibility. Purely additive. Before, memo instructions rendered via unknown_program; now they get a dedicated, richer rendering. No existing preset, dispatch order, or output changes — build.rs keeps the catch-all unknown_program ordered last, and the new visualizer only claims the memo program ID.

How it was built. Dispatch is driven by SolanaIntegrationConfig: the preset registers the memo program ID, and visualize_with_any selects MemoVisualizer ahead of the catch-all because build.rs orders specific visualizers first. The decode is a single std::str::from_utf8 with explicit fallbacks; there is no unwrap/expect/panic/unsafe (all denied by workspace lints).

Risks mitigated.

• Malformed or hostile input: invalid UTF-8 and empty data are handled explicitly and never panic; the raw hex is always shown so nothing is hidden from the signer.
• Spoofing: Memo Program is already a reserved canonical name and the program ID is in the trusted set in builtin_programs.rs, so a caller-supplied IDL cannot impersonate or override it.
• No existing fixtures render a memo instruction; the memo references in the jupiter/orca/meteora fixtures are account references handled by those programs' own visualizers, so no snapshot output changes.

Checks run (by agent)

• cargo fmt -p visualsign-solana — clean
• cargo clippy -p visualsign-solana --all-targets -- -D warnings — clean
• cargo clippy -p visualsign-solana --features diagnostics --all-targets -- -D warnings — clean
• cargo test -p visualsign-solana and cargo test -p visualsign-solana --features diagnostics — pass (8 new memo unit tests)
• make -C src test (full workspace + gRPC integration) — exit 0, 0 failures
• End-to-end: built a real memo VersionedTransaction and ran it through parser_cli --chain solana --output human; confirmed it dispatches to MemoVisualizer and renders the decoded memo text. The throwaway harness used to mint that transaction was removed before commit.

Manual steps needed (by human)

• None — fully automated by CI.

4. How this is maintainable

• The preset is self-describing: a module doc comment explains why Memo uses the native pattern instead of the IDL scaffold, and points at compute_budget / associated_token_account as siblings.
• Auto-registration: build.rs discovers MemoVisualizer from the directory name, so there is no hand-edited registry — adding or removing a preset is a directory drop-in.
• Rendering logic is split into a pure render_memo(program_id, data, index) function so it is unit-testable without constructing a VisualizerContext. The 8 tests cover UTF-8/unicode/empty/invalid decode, preview structure, the 1-indexed instruction label, config dispatch, and kind().
• A developer touching this needs to read only this one module; the field builders and the InstructionVisualizer/SolanaIntegrationConfig contract are identical to every other Solana preset.

🤖 Generated with Claude Code