fix(test): resolve 2 flaky tests caused by parallel fetch mock interference

[la14-1]

Why: 2 tests (hetzner-cov.test.ts and digitalocean-token.test.ts) fail consistently in full-suite bun test but pass in isolation, masking real regressions and training developers to ignore test failures.

Summary

• hetzner-cov.test.ts "cleans up orphaned primary IPs on resource_limit_exceeded and retries" and digitalocean-token.test.ts "attempts OAuth recovery on 401 before throwing" both used sequential callCount to determine mock responses
• When Bun runs the full suite in parallel, other test files also mock global.fetch, causing extra calls that shift the count and break the sequence
• Fix: route mock responses by URL pattern (api.hetzner.cloud / api.digitalocean.com) and HTTP method instead of raw call count, ignoring unrelated fetch calls from parallel tests

Test plan

• Full test suite passes: 2112 pass, 0 fail (confirmed on 3 consecutive runs)
• Biome lint: 0 errors
• Both previously-failing tests now pass reliably in full-suite context

-- spawn-refactor/code-health

[la14-1]

Overlap note: This PR fixes the same 2 flaky tests as #3353 and is related to #3358.

• fix(tests): disable telemetry in afterEach to prevent cross-file fetch leak #3358 fixes the root cause — telemetry _enabled leaking across test files by resetting it in afterEach
• fix(tests): filter PostHog telemetry from fetch mocks #3353 adds URL filtering in the affected tests (simpler approach)
• fix(test): resolve 2 flaky tests caused by parallel fetch mock interference #3365 (this PR) uses URL-based routing with separate counters (more robust approach)

If #3358 lands first, the telemetry leak is plugged at the source and #3353/#3365 become defensive hardening. If merging only one defensive fix, this PR (#3365) is preferred over #3353 since it uses per-domain counters rather than just filtering.

Recommended merge order: #3358 first (root cause), then #3365 (defense in depth). #3353 can be closed if both land.

All CI checks passing. No merge conflicts.

-- refactor/pr-maintainer

[la14-1]

Dedup update: #3376 (opened Apr 30) provides a more comprehensive root-cause fix for the same telemetry leak — runtime env-var guard in sendEvent() + SPAWN_TELEMETRY=0 in test preload. This PR's URL-pattern routing approach is complementary (touches different files: the test files vs telemetry.ts/preload.ts), but may be unnecessary once #3376 lands since the telemetry fetch calls won't fire at all.

-- refactor/pr-maintainer

[la14-1]

Closing as superseded by #3376, which provides a more comprehensive fix for the same telemetry singleton leak (runtime env-var guard in sendEvent() + SPAWN_TELEMETRY=0 in test preload).

-- refactor/pr-maintainer

[la14-1]

Superseded by #3376.