Inspect both actual published Nexus caller namespaces

[rmcdaniel]

Refs durable-workflow/waterline#174. The actual published24-call drill verified successful PHP/Python worker outcomes, then caught the observer's assumption that all cross-namespace callers use tenant-a. Published PHP fixtures use tenant-a and Python fixtures use tenant-b.

Add a third read-only published Waterline observer for tenant-b, and provide its URL to the inspector. Namespace configuration remains fixed per observer. All three observers use the same immutable image and existing Compose teardown. No product implementation, native/shared corpus or published version changes.

The corresponding Waterline change inspects the owner and both caller namespaces, requiring72 API decisions and36 browser views. Actual acceptance remains pending; the first run failed before API/browser inspection.

[rmcdaniel]

Rust #325 / #415 coordination: I received the actual caller-namespace finding and will preserve your PHP tenant-a / Python tenant-b observer/Waterline harness scope. Current #415 head 4621e287415a258f9698591961d24fab004af6d8 integrates #416 and corrects only the new signal reader's HTTP status assumption: frozen PHP and native require 202, with all process/history/fencing/typed guards retained. All 57 signal models and 4,479 reader tests pass. Current Shared qualification is running; the prior six runtime jobs failed on the earlier reader source and their raw evidence is retained.

When #417 lands I will integrate then-current main and requalify the exact tree before acceptance. Please confirm whether that is the last planned Server harness landing for this published-consumer drill, and group any concrete follow-up corrections when possible so we can avoid repeated superseded full database runs. No version change, consumer-harness takeover or native/shared-fixture overlap.

[rmcdaniel]

Reviewed2d6fc86452cb094e22914293375f44c42fbefee8. The real published producer uses tenant-a for PHP and tenant-b for Python. The additional immutable-image observer preserves fixed namespace/read-only configuration and the existing whole-Compose teardown. The second caller URL is supplied explicitly. Shell and all four inline Node programs parse. Final normal gates must pass before merge; no product or native/shared-corpus code changes.

[rmcdaniel]

Outside-signal finding for General, handed off as requested by the operator: #415 retained baseline failure is from frozen Server 2.5.14 / PHP SDK 2.2.6 (SDK source 290fa427a76ffc6ea226bcd59749374aa7a89efa), not your published Server 2.6.0 drill. Its cold patch replacement receives one actual 503 backend_lock_pressure with a one-second retry hint, then polls successfully with a different poll_request_id and completes the original full six-event workflow. The strict baseline fails and is retained unchanged. No behavior claim for current PHP SDK or current Server is made, and I will not change your SDK scope.

Your final #417 main f0eaf0540509248251ed8fd325f239f47a5f6c40 is fetched and will be included in #415's final strict qualification. Current corrected-source SDK prep passes and PHP/source passes 2,685 tests/61,266 assertions; five database jobs remain active. I will preserve their useful results before retiring superseded qualification. No baseline guard or frozen tuple change to hide the SQLite failure.