feat(bdd): [agent]/[no-ai] bracket annotations replacing @-style comment markers (#2678)

* feat(bdd)!: bracket comment-marker syntax to avoid cucumber tag confusion

Replace the @-prefixed step-marker comments (# @agent, # @no-ai, # @Soft)
with square-bracket markers (# [agent], # [no-ai], # [soft]) so per-step
comment annotations can no longer be mistaken for cucumber's native tag
syntax. Real Gherkin tags (@flow, @param:, scenario-level @no-ai/@Soft)
are unchanged, and $skill tokens still work on marker lines.

Clean break (the package is unreleased): the old syntax no longer routes,
but collectAnnotationFootguns now flags marker-only @-style comments with
a migration hint, surfaced as a new 'legacy-annotation' health kind in
the explore model and dashboard.

* feat(bdd-dashboard): regenerate dev fixture for bracket marker syntax

Regenerated from packages/bdd/example via `midscene-bdd dashboard` after
the comment markers moved to # [agent] / # [no-ai] / # [soft]; picks up
the shifted line numbers from the gherkin-tour explainer comments.

* refactor(bdd): derive marker regexes from one source and trim legacy-marker copy

* chore(bdd-dashboard): tighten legacy-annotation health description

* refactor(bdd): structured annotation footguns per simplify review

collectAnnotationFootguns now returns {kind, line, message} findings, so
the explore model maps health kinds directly instead of sniffing warning
prose (a copy-edit could silently reclassify findings) and stops
re-parsing uri:line out of the message. Legacy @-markers at the start of
a prose line ("# @agent check the logs") now get the migration hint —
they routed before the syntax change, and silence was the one migration
gap. Marker names and the skill-token grammar each derive from a single
source, and the audit reuses the comment normalization markerBody
already computed.

---------

Co-authored-by: ScriptedAlchemy <zack@module-federation.io>

Author: ScriptedAlchemy

apps/bdd-dashboard/src/components/HealthView.tsx

@@ -40,12 +40,17 @@ const KIND_META: Record<HealthKind, KindMeta> = {
   'detached-annotation': {
     label: 'Detached annotation comments (ignored)',
     severity: 'warn',
-    desc: 'An annotation comment (# @agent, # @no-ai, …) is not directly above a step, so it silently has no effect.',
+    desc: 'An annotation comment (# [agent], # [no-ai], …) is not directly above a step, so it silently has no effect.',
+  },
+  'legacy-annotation': {
+    label: 'Legacy @-style markers (ignored)',
+    severity: 'warn',
+    desc: 'A comment uses the retired @-marker syntax (# @agent, …), which no longer routes — write the bracket form (# [agent], …) instead.',
   },
   'tag-level-agent': {
     label: 'Tag-level @agent (ignored)',
     severity: 'warn',
-    desc: '@agent only works as a comment directly above a step — as a tag on a scenario it is ignored.',
+    desc: 'Agent routing only works as a # [agent] comment directly above a step — @agent as a tag on a scenario is ignored.',
   },
   'unused-flow': {
     label: 'Unused flows',

apps/bdd-dashboard/src/components/HelpPanel.tsx

@@ -82,8 +82,9 @@ const GLOSSARY: [string, string][] = [
   [
     'Step routing',
     'Where a step executes. Unmarked steps run on the Midscene UI agent; an ' +
-      '"agent" badge (# @agent or a $skill token) sends the step to a ' +
-      'general coding agent; "no-ai" runs a classic user-registered callback.',
+      '"agent" badge (a # [agent] comment or a $skill token) sends the step ' +
+      'to a general coding agent; "no-ai" runs a classic user-registered ' +
+      'callback.',
   ],
   [
     'soft',