fix(SDK-6463): catch a11y afterEach failures (cy.on fail guard) + circuit-break repeated scan timeouts

[Bhargavi-BS]

SDK-6463 (follow-up) — catch accessibility afterEach failures + circuit-break repeated scan timeouts

Builds on sdk-6463-nx-tsconfig-and-a11y-afterEach-fixes (base of this PR). The always-settle rework there stops the scan/save promises from hanging, but the customer's report shows the hook can still fail from its own Cypress commands — and any hook failure aborts the rest of the spec:

• cy.window() failing on a cross-origin page (their SSO redirect to login.microsoftonline.com)
• cy.task('get_test_run_uuid') failing when the TO node-side plugin isn't registered
• any other command-level error inside the hook

The "catch" (Cypress-correct)

Cypress chains have no .catch — the native mechanism is the per-test fail listener. The hook now registers, first thing:

cy.on('fail', (err) => {
  console.warn(`BrowserStack Accessibility: suppressed afterEach error: ${err && err.message}`);
  return false;   // suppresses the failure — test not failed, spec not aborted
});

The listener is scoped to the current test and auto-removed afterwards. Accessibility bookkeeping can no longer fail a user's test.

Circuit breaker for a dead scanner

Each hung scan/save costs up to ACCESSIBILITY_SCAN_TIMEOUT (default 25s) per test — a dead scanner stalls every remaining test's afterEach and every wrapped command. After 3 consecutive timeouts (configurable via ACCESSIBILITY_SCAN_CIRCUIT_LIMIT) the plugin stops attempting accessibility work for the rest of the spec, logging one warning.

Verified with REAL Cypress 13.15.0 (headed, no mocks)

Scenario	Without this PR	With this PR
Unregistered get_test_run_uuid task (responsive scanner)	hook fails → "skipping all of the remaining tests" → 0 passing	all tests pass, failure suppressed
Hung scanner (A11Y_SCAN_FINISHED never fires)	(already tolerated by base branch)	all tests pass
Circuit breaker (4 tests, hung scanner)	scan attempted every test	exactly 2 A11Y_SCAN dispatches (server-side count), circuit opens, later tests not stalled — 4/4 pass

The suppression mechanism itself was validated standalone in real Cypress first: cy.on('fail', () => false) in an afterEach suppresses a genuine command failure without aborting the spec (and cy.wrap(...).then(...).catch(...) throws TypeError: ...catch is not a function, which is why a literal .catch is not possible).

Unit tests: +2 (fail-listener contract, circuit-breaker dispatch stop); suite has no new failures vs base branch (685→687 passing, same 16 pre-existing failures).

🤖 Generated with Claude Code