…reporter

The InternalTelemetrySystem ITS reporter was hard-coded to use
SendPolicy::default(), which has console_fallback: true. Under load
this can produce hundreds of thousands of 'Channel full, dropping
observed event' raw_error! lines on stderr per second when the ITS
flume channel saturates, which:
  - drowns out useful diagnostic output
  - silently dominates /proc/<pid>/io wchar so any wire-bytes
    measurement based on that counter is contaminated
  - costs measurable CPU writing those messages

The engine config already has the right knob,
engine.observed_state.logging_events, with default
console_fallback: false for the logging path (only the engine_events
path defaults to true). It just was not being plumbed to the ITS
reporter.

Add an its_reporter_policy: SendPolicy parameter to
InternalTelemetrySystem::new and use it for both reporter
constructions in the ITS branch (with and without log tap). Pass
engine.observed_state.logging_events.clone() from the controller
startup. All test call sites updated to pass SendPolicy::default()
to preserve their existing behaviour.

Verified with cargo test -p otap-df-telemetry --lib (126 passed).
Verified end-to-end: a logger configured with global: its at 10k/s
and a downstream that backpressures (forcing channel-full drops at
~50%) now produces 0 'Channel full' lines vs the prior 150k lines
in 30s, and process CPU drops correspondingly.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

…policy

Covers both code paths in InternalTelemetrySystem::new where the ITS
reporter is constructed (with and without an InternalLogTapHandle),
verifying that the configured engine.observed_state.logging_events
SendPolicy is threaded through rather than replaced with
SendPolicy::default().

Adds test-only accessors:
- ObservedEventReporter::policy()
- InternalTelemetrySystem::its_reporter()

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>