fix(observability): keep prometheus in profile set when enabling grafana sidecar

[srikanth-bitdynamics]

Problem

dpm localnet observability enable --name <inst> (and the Web UI Metrics "Enable observability now" button, which shares the orchestration) fails on a running instance with:

observability toggle failed: enable grafana: docker compose up: exit status 1
output:
service "grafana" depends on undefined service "prometheus": invalid compose project

Reproduced on Splice 0.6.4, Compose v5.0.2, on a healthy instance created without --profile observability.

Root cause

SetObservability brings up prometheus and grafana in two separate docker compose up invocations, each activating only its own per-component profile. The grafana step ran --profile grafana up -d grafana, but the overlay (assets/compose/observability.yaml) gates prometheus behind profiles: ["observability","prometheus"] and grafana declares depends_on: prometheus. With only the grafana profile active, Compose filters prometheus out of the project model, so grafana's depends_on references an undefined service and the whole project is rejected. This bit even the enable-both path: prometheus comes up fine, then the separate grafana invocation fails.

Fix

Whenever grafana is started, activate both --profile prometheus --profile grafana so prometheus is a defined service (satisfying depends_on), and scope the up with --no-deps so activating that profile does not drag a prometheus container up in the legitimate grafana-only / external-scrape case (which the existing warning documents).

• New observabilityUpProfiles + observabilityUpArgs helpers build the argv; enableObservabilitySidecar drops its now-redundant profile param.
• Kept the overlay's depends_on — it still orders startup in the umbrella up.go path.
• Both the CLI verb and the Web UI Metrics toggle go through the shared SetObservability, so CLI ↔ Web UI parity holds with no per-surface docker drift.

Verified scenarios: grafana onto an already-running prometheus, both from cold, and grafana-only (prometheus defined-but-not-started, warning still surfaced).

Test

TestObservabilityUpArgs_GrafanaKeepsPrometheusInProfileSet materializes the real overlay (injecting a bytes.Buffer for the drift writer per the repo testing rule) and asserts the generated grafana up argv keeps prometheus in the active profile set and passes --no-deps, while a prometheus-only up does not activate the grafana profile.

make test and make lint are green.

Related (not fixed here)

The same root cause also affects cold-start dpm localnet up --profile grafana (grafana-only): up.go runs a whole-project up under --profile grafana with prometheus's profile inactive, producing the identical rejection. That path can't use the toggle's --no-deps trick (no explicit service list), so the clean fix there is to drop grafana's depends_on and rely on runtime datasource discovery — a separate change to the overlay's dependency contract, left out of this scoped fix.