[CCOR-13193] Run Integration(e2e) tests on PR-to-main (phase 1: test with sdkdev v5 server)

[chrishagglund-ship-it]

Summary

Enables the integration test suite to run in CI on pull requests and hardens the
suite to run reliably against the shared sdkdev dev server. Introduces a shared
retry/poll helper module, splits the slowest tests into parallel CI buckets, and
fixes a number of latent test bugs surfaced by actually running these tests in CI.

Production code impact is limited to two small, backward-compatible model additions.

Changes

CI

• Add an integration-test job to pull_request.yml running four parallel buckets
  (test-all, long-sync, long-async, core) via a matrix, with a 30-min timeout.
• Trigger CI on pull_request/push to main (plus workflow_dispatch); scope the
  concurrency group by head_ref so PR runs cancel their own in-progress runs.
• Force HTTP/1.1 for integration tests (CONDUCTOR_HTTP2_ENABLED=false) to avoid
  intermittent mid-stream HTTP/2 stalls through the shared server's proxy/LB.
• Move integration credentials to the integration job only; unit/bc/serdeser steps
  no longer request server secrets (they run offline).

Test infrastructure

• Add tests/integration/retry_helpers.py: the single home for transient-blip
  detection (is_transient), per-request retry (retry_on_transient,
  retry_on_status), scenario-level retry with a shared wall-clock deadline
  (retry_scenario), and a shared poll-to-terminal primitive
  (wait_for_workflow_terminal).
• Add scripts/run_integration_tests.sh to run the suite locally, mirroring the CI
  buckets (AI/agentic and perf tests excluded by default; --bucket= / --with-perf).
• Add pytest markers (slow_sync, slow_async, slow_test_all,
  server_timeout_unreliable) in pyproject.toml to drive bucket selection.

Reliability fixes in tests

• Suffix task/workflow names with a per-run id so concurrent runs on the shared
  server don't steal or strand each other's tasks.
• Start workers once per class via a TaskHandler stopped in tearDownClass
  (replacing a daemon thread + fixed sleep(90) that could hang the interpreter
  at exit), and assert each expected worker is discovered and alive.
• Retry idempotent reads on read-after-write 404s and workflow terminate on 423 Locked;
  add server-side diagnostics (queue sizes, pollers, worker liveness) on give-up.
• Fix genuine test bugs: access WorkflowSchedule as an object not a dict; drain the
  shared task queue by each polled task's own workflow id/ref; add a missing assert
  on task logs; correct an access-key-lookup assertion.
• Rename test_* scenario helpers that take a workflow_executor arg to scenario_*
  so pytest stops mis-collecting them as standalone tests.

Models (backward-compatible)

• UpsertUserRequest: add optional contact_information: Dict[str, str] field.
• ServiceMethod.example_input: dict -> object.

Risk

• Production surface is limited to two additive model changes; verified against the
  existing backward-compatibility and serde tests.
• Remaining changes are test/CI only.

Notes for reviewers

• CI now only runs for PRs targeting main (previously all PRs). Confirm this matches
  the branching model.
• The server_timeout_unreliable no-heartbeat timeout cases are excluded from CI
  buckets (the dev server doesn't fire the timeout on a CI-bounded timeline); the
  SDK heartbeat-keeps-alive path is still covered.

[codecov]

Codecov Report

❌ Patch coverage is 75.00000% with 4 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...onductor/client/http/models/upsert_user_request.py	75.00%	4 Missing ⚠️

Files with missing lines	Coverage Δ
src/conductor/client/http/models/service_method.py	58.13% <ø> (ø)
...onductor/client/http/models/upsert_user_request.py	85.41% <75.00%> (-2.54%)	⬇️

... and 3 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.