fix(lint): use string comparison to dodge GHA == coercion

[topcoder1]

Summary

Critical fix for the reusable lint workflow's input-driven skip logic. Discovered while watching the install-fanout PRs (wxa-jake-ai#181 + wxa_vpn#180): both set `run_prettier: false` in the caller, but the prettier job ran anyway.

Root cause

GHA's `==` operator does loose-numeric coercion. `null` and `false` both coerce to `0`, so `false == null` evaluates to true. The previous condition

```yaml
if: ${{ inputs.run_prettier == null || inputs.run_prettier }}
```

was supposed to mean "run on self-test (input null) OR when explicitly true." But when the caller passed `run_prettier: false`, the `== null` clause matched (false coerces to null!) and the job always ran.

Fix

`format('{0}', inputs.X)` returns:

• `''` when input is null (self-test)
• `'true'` / `'false'` for explicit booleans

So:

```yaml
if: ${{ format('{0}', inputs.run_prettier) != 'false' }}
```

Context	format result	!= 'false'	Behavior
Self-test (pull_request)	`''`	true	Run ✓
workflow_call run_X: true	`'true'`	true	Run ✓
workflow_call run_X: false	`'false'`	false	Skip ✓

Applied to:

• actionlint job-level `if:` (run_actionlint)
• prettier job-level `if:` (run_prettier)
• install-deps step-level `if:` (install_node_deps)
• shellcheck branch inside actionlint step (run_shellcheck)

After this lands

wxa-jake-ai#181 and wxa_vpn#180 will correctly skip the prettier job on next CI re-run.

🤖 Generated with Claude Code

[claude]

No issues found. The format('{0}', inputs.X) != 'false' pattern is correct for all three input states (null → '', true → 'true', false → 'false'), and the shellcheck branch rewrite is semantically equivalent to the old ternary for all valid boolean inputs.