Surface MPS control-daemon impairment reason as StatusReason

[harishxr]

Summary

Surface a human-readable reason alongside the instance health status. On IMPAIRED, the MPS control-daemon health check now reports why e.g. "probe timed out, daemon wedged" or "probe failed, cannot find MPS control daemon process", which reaches DescribeContainerInstances and the "ECS Container Instance Health Change" EventBridge event as statusReason. Follow up to #5124, which reported status only.

The plumbing is generic and any doctor health check that returns a non-empty reason gets it forwarded to the wire. Only the MPS check populates one in this change.

Implementation details

The doctor Healthcheck interface gains a GetStatusReason() accessor and SetHealthcheckStatus now takes the reason alongside the status so the two are always set together. The shared status tracker stores and returns the reason under its existing lock, so every check that embeds the tracker picks up both new methods for free.

On the publish side, the TCS client forwards a non-empty reason to InstanceStatus.StatusReason and leaves the pointer nil when there is none, so omitempty drops it and the backend sees null rather than an empty string.

Only the MPS control-daemon check populates a reason today: on IMPAIRED it reports the probe error (daemon wedged / probe timed out / control binary missing), bounded to MHS's 1024-character limit on a rune boundary, and clears it on any healthy tick. Every other check stays status-only, so their behavior is unchanged; the plumbing is generic, so a future check can surface a reason by simply returning one.

Testing

New tests cover the changes: yes

Agent built from this branch:

• Baseline: DescribeContainerInstances --include CONTAINER_INSTANCE_HEALTH → ACCELERATED_COMPUTE = OK, no statusReason; CONTAINER_RUNTIME = OK.
• systemctl stop nvidia-mps.service: after the 3-strike threshold (~70s), ACCELERATED_COMPUTE = IMPAIRED with statusReason = "mps control daemon probe failed (exit 1, output "Cannot find MPS control daemon process"): exit status 1". systemctl start → back to OK, statusReason cleared.
• Wedge path (kill -STOP the control daemon so the probe times out): after ~70s, ACCELERATED_COMPUTE = IMPAIRED with statusReason = "mps control daemon probe timed out after 3s (daemon wedged?): signal: killed" (timedOut=true in the log). kill -CONT → OK, statusReason cleared.
• Both channels, both paths: the reason appears identically on DescribeContainerInstances --include CONTAINER_INSTANCE_HEALTH and on the "ECS Container Instance Health Change" EventBridge event (ec2InstanceId set, capacityProviderName: null), and clears on recovery on both.

Description for the changelog

Enhancement - Surface MPS control-daemon impairment reason as StatusReason

Additional Information

Does this PR include breaking model changes? If so, Have you added transformation functions? No

Does this PR include the addition of new environment variables in the README? No

Licensing

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.