feat(reporting): align delegated reporter protocol with ReportingContext and multi-type dispatch

[saurabh-net]

Context & Problem

This PR aligns the delegated remote reporting mechanism (RemoteReporter) with the same structured context model established for delegated scorers:

1. Structured ReportingContext in Protobuf:
   Previously, ReportingRequest only contained ReporterSpec (name, configuration JSON, timeout), omitting evaluation context (job_id, run_time, store_type, DataFrame records, and database). We introduce ReportingContext in eval_agent.proto:

   message ReportingContext {
     string job_id = 1;
     string run_time = 2;          // Timestamp of evaluation run
     string store_type = 3;        // "CONFIGS", "EVALS", "SCORES", "SUMMARY"
     string results_json = 4;      // Serialized DataFrame records as JSON string (orient="records")
     string database = 5;          // Optional target database name
   }
   
   message ReportingRequest {
     ReporterSpec reporter = 1;
     ReportingContext context = 2;
   }

2. Multi-Type Dispatch (Option 2):
   Native EvalBench reporters (Reporter.store) are invoked across four distinct store phases: STORETYPE.CONFIGS, STORETYPE.EVALS, STORETYPE.SCORES, and STORETYPE.SUMMARY. Previously, RemoteReporter had hardcoded filtering (if type_name != "EVALS": return), which prevented remote reporters (like CSV or custom sinks) from receiving and persisting other report tables. We now dispatch every store_type across the stream along with its serialized DataFrame rows, allowing the remote worker to decide which types to process.

3. Logging & Diagnostics:

   • Standardized remote reporting logs with clear [REMOTE_REPORTER] prefixes for easy debugging across stream sessions.
   • Clean handling of timeouts and failure responses without crashing the evaluation runner.

Changes

• evalbench/evalproto/eval_agent.proto: Added ReportingContext message and updated ReportingRequest to include context.
• evalbench/reporting/remote_reporter.py: Constructed ReportingContext, serialized DataFrame records with ISO date formatting, dispatched across all store_type calls, and handled response logging.
• evalbench/test/agent_grpc_proxy_test.py: Added comprehensive unit tests covering ReportingContext field validation, multi-type dispatch across all 4 store phases (CONFIGS, EVALS, SCORES, SUMMARY), timeout handling, and failure response handling.

Verification

• pytest evalbench/test/agent_grpc_proxy_test.py evalbench/test/agent_grpc_proxy_integration_test.py (9 passed, 0 warnings).
• pycodestyle clean (0 errors).
• Zero generated proto stubs in diff against main.

[prernakakkar-google]

/gcbrun