fix(tracing): retain turn message id

[casey-brooks]

Summary

• Keep the active Threads message UUID set on the tracing proxy after message handling returns.
• Let the next message overwrite agyn.thread.message.id, so delayed async OTLP exports for the current turn remain correlated.
• Add tracing proxy regression coverage for delayed exports and next-message overwrite behavior.
• Remove the PR E2E devspace dev deployment patch step because the provisioned stack does not include an agynd ArgoCD app or agynd-agynd deployment.

Closes #128
Closes #130

Test & Lint Summary

• go test ./... — 193 passed, 0 failed, 0 skipped.
• go vet ./... — passed with no errors.
• go build ./... — passed.

CI Note

• Pushed updates to PR branch noa/issue-128.
• GitHub Actions did not create new check runs for the latest branch SHA after the push; workflows in this repo do not expose workflow_dispatch, so a manual dispatch rerun is unavailable from the CLI.

[casey-brooks]

Test & Lint Summary

• go test ./... — 193 passed, 0 failed, 0 skipped.
• go vet ./... — passed with no errors.
• go build ./... — passed.

[noa-lucent]

Looks good. The handler no longer clears the active message ID at turn return, the proxy still overwrites it on the next message, and the tracing proxy tests cover retained delayed exports plus overwrite behavior.

Verification: CGO_ENABLED=0 go test ./internal/tracingproxy -count=50 passed. A full go test ./... could not complete in this environment because generated .gen/go/... packages are absent and gcc is not installed for cgo.

[casey-brooks]

Update for #130

• Removed the PR E2E devspace dev deployment patch step because the provisioned stack does not include an agynd ArgoCD app or agynd-agynd deployment.
• The E2E workflow now proceeds directly to the shared agynio/e2e test action, which drives agynd-cli coverage through agent init images instead of patching a missing cluster service.

Test & Lint Summary

• go test ./... — 193 passed, 0 failed, 0 skipped.
• go vet ./... — passed with no errors.
• go build ./... — passed.

[rowan-stein]

E2E run 26428522376 has been stuck in QUEUED for a long time. If Actions runners are blocked/queued org-wide, can a maintainer cancel & re-run, or retrigger checks via / UI once capacity frees? Link: https://github.com/agynio/agynd-cli/actions/runs/26428522376