feat(surfpool,skill): auto-discover preset IDLs and fix solana-add-idl template gaps

[shahan-khatchadourian-anchorage]

Summary

Wires the surfpool harness to cover skill-generated presets via reflection, and fixes three correctness gaps in the solana-add-idl skill that would have produced code that does not compile.

Stacked on top of #285 (shahankhatch/surfpool-rust-runner).

Reflection — surfpool now auto-discovers preset IDLs

• build.rs scans src/presets/<name>/<name>.json and emits pub const PRESET_IDLS: &[(&str, &str)] to OUT_DIR; lib.rs re-exports it. Dropping a JSON file at the matching path makes it appear on the next build — no manual registration.
• tests/surfpool_fuzz.rs::surfpool_preset_idls iterates PRESET_IDLS and runs each through the same roundtrip used for the named upstream-IDL tests. Currently empty (no preset ships an embedded IDL today); will pick up the first skill-generated preset.
• run_idl_roundtrip + idl_test! macro hoisted from surfpool_fuzz.rs into tests/common/mod.rs so future per-preset test files can reuse them. #[macro_export] + $crate::common::run_idl_roundtrip path makes the macro work from any test that does mod common;.

Skill fixes — .claude/skills/solana-add-idl/SKILL.md

• Removed the squads_multisig template reference — that preset does not exist in this repo. Replaced with unknown_program, which is the actual working IDL-parsing reference.
• Removed references to five non-existent solana_parser helpers (build_named_accounts, build_parsed_fields, build_fallback_fields, append_raw_data, format_arg_value). Documented the manual named_accounts loop pattern that unknown_program uses.
• Added explicit step for declaring pub(crate) const {SCREAMING_SNAKE}_PROGRAM_ID: &str in the preset's mod.rs — the generated config.rs references it via use super:: but the previous skill never told contributors where to put it.
• Removed the idl_test! registration step (replaced by PRESET_IDLS auto-discovery) and added a verification command that greps the generated preset_idls.rs.
• Documented what's already auto-covered (proptest is generative, cargo-fuzz is generative, surfpool roundtrip is reflective) vs. what still needs hand-written assertions (semantic_pipeline.rs).

Test plan

• cargo fmt -p visualsign-solana
• cargo clippy -p visualsign-solana --all-targets -- -D warnings
• cargo test -p visualsign-solana — 111 unit/integration tests pass; 15 surfpool tests #[ignore] (the new surfpool_preset_idls discovers as the 15th)
• cargo test -p visualsign-solana --test surfpool_fuzz -- --list lists surfpool_preset_idls alongside the existing 14 named upstream-IDL tests
• OUT_DIR/preset_idls.rs is generated with pub const PRESET_IDLS: &[(&str, &str)] = &[ ]; (empty — no preset has an embedded IDL JSON yet, which is the expected initial state)
• Surfpool label run on this PR — exercises surfpool_preset_idls (will be a no-op until a preset ships an IDL JSON, but confirms the test discovers cleanly)

🤖 Generated with Claude Code

[shahan-khatchadourian-anchorage]

Coordination notes — overlap with #278 and #281

Surfaced during a follow-up triage. Posting here so reviewers and downstream rebases know what to expect.

Overlap with PR #278 — auto-generate presets/mod.rs

#278 (build(solana,sui): auto-generate presets/mod.rs from filesystem) solves the same problem this PR's commit a45e514 solves, with the same #[path] mechanism, but with broader and better-engineered scope:

• Covers both visualsign-solana and visualsign-sui (this PR only covers solana)
• Rewrites build.rs to be Result-returning (no unwrap/expect, no #[allow(clippy::unwrap_used)] dodge — properly aligned with the workspace lint policy in CLAUDE.md)
• Motivated by quantified pain: 12 in-flight preset PRs colliding on the alphabetical pub mod list during rebase

Suggested resolution: drop commit a45e514 from this PR (revert the auto-mod portion) and rebase on top of #278 once it lands. The PRESET_IDLS reflection (pub const PRESET_IDLS: &[(&str, &str)]) and the SKILL.md scope refresh stay; only the presets/mod.rs auto-generation work goes back to #278.

Overlap with PR #281 — skill refresh for post-#228 API

#281 (chore(skill): refresh solana-add-idl for post-#228 VisualizerContext API) is a parallel refresh of the same skill, against a different stack:

• It's stacked on shahankhatch/228-lint-diagnostics-refactor, where context.current_instruction() is being removed in favor of context.resolve_program_id()? / context.resolve_accounts()? / context.data().
• It points at dflow_aggregator as the real working template (replacing the dead squads_multisig reference) — the same fix this PR makes by pointing at unknown_program.
• This PR's SKILL.md tells contributors to copy the named_accounts loop from unknown_program::try_parse_with_idl — that loop uses context.current_instruction() and will compile-rot the moment feat: attested transaction lint diagnostics in SignablePayload #228 lands.

Suggested resolution: #281 lands first, then this PR's SKILL.md changes rebase on top of it. The pieces of this PR's SKILL.md edits worth carrying forward post-rebase:

• The new "Scope" section (lines 13-36) framing the skill as structural-not-semantic
• Step 5's promotion of the tests/semantic_pipeline.rs note to its own subsection
• Step 8 "What's next — semantic refinement (optional, follow-up)"
• The {SCREAMING_SNAKE}_PROGRAM_ID declaration step

Most of the import list and parse_instruction_with_idl-pattern guidance overlaps with #281 and should be reconciled at rebase time.

Out of scope here

• The three triage targets (CI: Detect changes to on-chain IDL for Solana and fail visualizations if changed #181, Solana IDL based Program Implementation Roadmap #182, Split visualsign-solana binary #185) are thin placeholders — CI: Detect changes to on-chain IDL for Solana and fail visualizations if changed #181 depends on a chain (Solana dApp Testing Framework #180 → Add a tool to extract unsigned transactions from onchain explorers for solana #177) of unimplemented work, Solana IDL based Program Implementation Roadmap #182 is a mega-issue, Split visualsign-solana binary #185 is a one-line stub. Picking any up requires scoping first, not code yet.

tl;dr

This PR ships in two halves: a build.rs auto-mod commit that duplicates #278, and a SKILL.md rewrite that conflicts with #281. The clean path is to land #278 and #281 first, then this PR carries only PRESET_IDLS reflection + the SKILL.md pieces #281 doesn't already do (Scope section, Step 8, semantic-refinement handoff).