fix(trust): ignore self-reports in ScoreAverageTrust

Description

The pitch.

## Problem

`ScoreAverageTrust.report()` never checks `evidence.reporter`, so an agent can
file positive evidence about itself to inflate its own score, or dilute genuine
negative reports back toward neutral.

## Change

- `report()` ignores evidence where `evidence.reporter == agent`. All other
  reports behave as before. Matches `bonded_trust`, which already skips
  self-reports when scoring.
- Documented in the method docstring.

## Tests

- `test_self_report_is_ignored`: ten positive self-reports leave the agent at
  the never-rated default. Fails on main (score=1.0, sample_count=10).
- `test_self_report_does_not_dilute_others`: a self-boost cannot offset a
  genuine negative report. Fails on main (score=0.5, sample_count=2).

## Verification

`make ci-local` passes all five gates (pytest: 1276 passed, 1 skipped).