fix(psi-pro): correct two invalid psi.mode examples + test every example

[travisccook]

Found by strengthening the examples round-trip test to check all examples, not just the first (the gap that also hid the negative-int engine bug).

psi.mode (template {address}T{mode}, both enums) had two examples referencing codes not in the psi.mode enum:

• 5T03 — the code is 3, not 03 → 5T3 (Rear, Slow Flash)
• 0T6 — there is no mode 6 → 0T1 (All, Reset — preserves the address-0 'All' demo with a valid mode)

Both now round-trip to a structured, editable step. Also strengthens the round-trip test to exercise every example, and bumps libraryVersion 2.2.0 → 2.2.1 (manifest + releases.json + test assertions in sync). 137 tests pass; validator green.

[github-actions]

Review

Checks — all green.

• npm ci — clean install (272 packages, no errors; only pre-existing npm audit advisory, unrelated).
• npm run validate — passes for all 9 boards + cross-file check, including boards/psi-pro.json.
• npm test — 7 suites / 137 tests pass, including the newly strengthened test/web.test.js round-trip check.

What this PR does. Pure data fix in libraries/boards/psi-pro.json:104-108: two psi.mode command examples referenced enum codes that don't exist —5T03 (valid code is "3", not "03"; there's no zero-padding on this param) and 0T6 (there's no "6" in the psi.mode enum at libraries/boards/psi-pro.json:21-65, which only has 1,3,5,9,11,13,15,17,18,92). Fixed to 5T3 (Rear / Slow Flash) and 0T1 (All / Reset), both valid.

Board-contribution rules — all satisfied:

• examples present (unchanged, still 3 per command). ✅
• Placeholder ↔ param mapping for {address}T{mode} untouched by this PR and still valid. ✅
• param.enum refs (psi.address, psi.mode) resolve; unaffected by this change. ✅
• Command ids unaffected (no new/renamed commands). ✅
• libraries/manifest.json bumped 2.2.0 → 2.2.1 (correct patch bump per semver rules — this is a data/example fix, not a new command or breaking rename). releases.json updated in lockstep (version + notes) and test assertions in test/engine.test.js, test/load-node.test.js follow suit. ✅
• confidence: "high" on psi-pro is pre-existing and untouched by this diff — out of scope here, not a regression.

Safety-class check: psi.mode (libraries/boards/psi-pro.json:88) is an LED-effect display command (safety: "cosmetic") — correctly cosmetic, not movement/power. No mislabeling introduced by this PR. No src/*.js changes in this diff, so the engine/UI/round-trip invariant is untouched.

Nice catch, worth calling out: the real value-add here is test/web.test.js:56-63 — the round-trip example test previously only checked cmd.examples[0], silently ignoring examples 2 and 3, which is exactly how these two bad codes went unnoticed. The PR changes it to loop over every example. This is a good, minimal, low-risk hardening of CI that will catch this class of bug in future board PRs — no concerns with the approach.

No issues found. LGTM from a data/process standpoint.