feat(plugins): add on_agent_error_callback and on_run_error_callback lifecycle hooks

[STHITAPRAJNAS]

Summary

Fixes #4774

When an unhandled exception propagates out of an agent's _run_async_impl / _run_live_impl, or out of the runner's execution loop, the existing after_agent_callback / after_run_callback were silently skipped. This made fatal failures invisible to observability plugins (e.g. BigQuery analytics), inflating success rates and losing failure events entirely.

Changes

• BasePlugin: add on_agent_error_callback(agent, callback_context, error) and on_run_error_callback(invocation_context, error) with safe no-op defaults.
• PluginManager: add run_on_agent_error_callback / run_on_run_error_callback dispatch methods backed by a new _run_error_callbacks helper that fans out to all plugins (no early-exit) and logs — but does not propagate — individual plugin failures.
• base_agent.py: wrap run_async / run_live generator loops in try/except; call run_on_agent_error_callback before re-raising.
• runners.py: wrap the execute_fn generator loop in try/except; call run_on_run_error_callback before re-raising. after_run_callback is intentionally skipped on the error path so plugins can distinguish clean completions from fatal failures.

Design decisions

Decision	Rationale
Error callbacks are fire-and-forget (no early-exit, no return value)	They are pure observers — a broken plugin must not prevent others from recording the failure
after_run_callback is not called on error	Allows plugins to distinguish success from failure; analogous to how on_model_error_callback vs after_model_callback work
Individual plugin failures are logged, not re-raised	A broken observability plugin must never hide the original error
The original exception is always re-raised	The framework does not suppress errors; this is notification only

Testing plan

• tests/unittests/plugins/test_lifecycle_error_callbacks.py — 12 tests covering BasePlugin defaults, PluginManager fan-out semantics, argument forwarding, and no-early-exit behaviour
• tests/unittests/runners/test_runner_error_callbacks.py — 7 integration tests covering runner error/success paths with real InMemoryRunner
• tests/unittests/agents/test_agent_error_callbacks.py — 11 tests covering run_async and run_live error paths
• All 688 existing plugin + agent + runner tests pass unchanged

[rohityan]

Hi @STHITAPRAJNAS , Thank you for your contribution! We appreciate you taking the time to submit this pull request.
Can you please fix the failing formatting tests. You can use autoformat.sh

[rohityan]

Hi @Jacksunwei , can you please review this.

[STHITAPRAJNAS]

Formatting done

[STHITAPRAJNAS]

Hi @Jacksunwei — gentle ping! This PR has been ready for review since Mar 25 with all checks passing. Would appreciate your feedback when you get a chance. Thanks!

[STHITAPRAJNAS]

Hi @Jacksunwei — following up on this. The on_agent_error_callback and on_run_error_callback hooks fill a real gap for anyone running plugins that need to distinguish clean completions from fatal failures (e.g. alerting, metrics, cleanup). The implementation is tight and the test coverage is solid. Would appreciate your thoughts whenever you have bandwidth.