feat(observability): add WH_OTEL_ENABLED to short-circuit telemetry init #96

Problem

In cmd/wavehouse-{api,worker}/main.go (~lines 74 / 54) and
internal/observability/provider.go, when WH_OTEL_ADDR resolves but no collector
is listening, the OTLP exporter logs traces export: exporter export timeout: rpc error: code = Unavailable … connect: connection refused on every flush interval
(plus quieter noise on the metrics side). During a phased rollout — WaveHouse up
before the collector/receiver is deployed — this spams logs continuously. There's no
off switch; the only workaround is pointing WH_OTEL_ADDR at an unreachable
address, which just trades the timeout warning for a retry warning.

Proposed Solution

Add WH_OTEL_ENABLED (default true). When false, observability.InitProvider
returns a no-op shutdown and skips both exporter setup and runtime.Start.
Operators running WaveHouse without a collector shouldn't have to run a sink to
silence boot.

Alternatives Considered

• Auto-detect collector reachability and disable on failure — racy, and hides
  genuine misconfiguration once a collector is expected.
• Just lower the exporter log level — loses visibility when telemetry should be
  working; doesn't avoid the wasted export attempts.

Additional Context

From nas-observability WHissues.md, 2026-05-08. Related to #47 (env-var config)
for the override-precedence convention. This is the missing off-switch for the OTLP
pipeline stood up by #14 (Centralize Telemetry via Single OTLP Pipeline) and #28
(OTLP Log Exporting) — both closed, neither added a way to disable export.