Improve OTel tracing for analytics and aws cli

[carole-lavillonniere]

Motivation

Traces emitted under LSTK_OTEL=1 were not fine-grained enough around the lstk aws wrapper. Investigation revealed several blind spots: the telemetry call, the Docker daemon-ID lookup used to derive machine_id, and the aws CLI subprocess itself were either uninstrumented or producing orphaned/dropped spans.

Changes

• Instrument the telemetry HTTP client with otelhttp.NewTransport
• Instrument the Docker client used by LoadOrCreateMachineID
• Wrap awscli.Exec in a span (aws cli)
• Reorder defers in cmd/root.go:Execute so tel.Close() runs before the tracer-provider shutdown (otherwise the analytics POST span is created on a noop tracer after shutdown and never reaches the exporter)
• Make machine-ID resolution lazy (in GetEnvironment(ctx)) so the Docker info / _ping spans are parented to the active command span instead of orphaned at process init.

toward DRG-809

Result of running lstk aws sts get-caller-identity:

[anisaoshafi]

LGTM 👏🏼

[anisaoshafi]

Question: there is a difference in time for lstk aws sts get-caller-identity from before to after 700 ms -> 1.98 s. Is that perhaps related to the improvement itself or just random?

[carole-lavillonniere]

Question: there is a difference in time for lstk aws sts get-caller-identity from before to after 700 ms -> 1.98 s. Is that perhaps related to the improvement itself or just random?

Good observation! I did some research and testing, and figured I was pointing to aws cli v1 which is in Python (v2 is in go hence a single binary). aws cli v1 is much slower and needs to do I/O when cache is not warm. The slowness does not seem to be due to anything added in this PR.