feat(ci): modernize ADO integration test pipeline

[Hangyi (HangyiWang)]

Since we're reusing ADO pipeline https://dev.azure.com/azureiotdevxp/aziotcli/_build?definitionId=147&_a=summary for FIC migration, this PR modernizes pipeline 147 / Azure IoT CLI - Integration Tests for DPS, HubControl, HubData, ADU and ADR, using the existing ADO service connection with renewable WIF and a simple Dry run / Integration tests form with Python, region and ARM-routing selection. The form defaults to Integration tests with the named DPS/Hub/ADR/ADU preset; clear the preset to select individual suites, or explicitly choose Dry run for planning only. Shipped CLI behavior, pipeline default branch, GitHub release/schedule callers and FICs are unchanged.

Build, Lint and Unit run in parallel, followed by parallel selected integration services against one immutable candidate wheel, with manual same-run recovery that preserves original results. Completed call failures rerun only the failed cases, including service/transport exceptions. When exact selection is unsafe or unavailable, the next manual attempt reruns the entire affected service/Python/region combination with fresh fixtures, including passing authentication phases. This covers fixture failures, incomplete execution, worker crashes, timeouts and installation/login failures. Full retries replace effective service results rather than hiding new failures behind old passes. Commit, candidate wheel, dependency fingerprint, target, ancestry and raw-artifact integrity remain enforced; changed immutable evidence requires a new run. Rejected attempts publish explicit failure reports, completed sibling receipts remain available, and startup gaps cannot reuse an older pass. Runs pinned to the older attempt format need a new run to use this policy.

ADO cleanup uses submission handoff rather than waiting for resource absence where supported. Accepted deletion is recorded separately from confirmed absence. Pending dependencies and failed submissions retain exact IDs in cleanup-status.json and ownership receipts for later operational cleanup, as requested; no extra cleanup job or new cleanup polling deadline is added. CSR targets stay retained while their namespace/device cleanup is unresolved. Pending resources may continue consuming quota. Ownership checks, worker termination and actual deletion assertions inside tests remain intact, and no generic ADR/ADU HTTP or inventory guard is introduced.

ADO runs are no longer blocked by other project builds or a fixed cleanup-time window; the separate GitHub live-run check remains. Standalone controllers and their subprocesses inherit candidate dependencies, with main/DPS interpreter startup checks before Azure login. ADU has no added per-test timeout, and DPS regular/service-SAS attempts retain seven-worker execution with worker-safe result collection; existing job/phase and process-shutdown limits, serial local-auth toggling and non-ADO cleanup requirements remain intact.

Depends on #917. The reusable unit shard planner, pytest plugin, duration profile, completeness/coverage gate, GitHub precheck wiring, and mocked Digital Twins polling speedup now belong to that standalone PR and are excluded from this diff. This PR retains the ADO-specific four-shard Unit/UnitGate adapter, pipeline artifact wiring, and adapter regression checks. ADO cross-job pip caching remains disabled after measurement showed no net benefit.

Merge order: merge #917 into release/1.1.0-preview first, then retarget this PR from users/hangyiwang/unit-test-parallelism to release/1.1.0-preview. If the unit PR is squash/rebase merged, also merge the updated release base into this migration branch to preserve the clean separation. Both branches have been rebased onto the release 1.1 stack; this PR remains based on the unit branch to keep its ADO-only diff.