fix: match dev server webhook payload to cloud format

[Swigler]

Summary

• Dev server webhook payloads now match the cloud format — includes id, status, input, webhook, delayTime, and executionTime fields alongside output/error
• Webhooks now fire on failed jobs — previously the error path returned before reaching the webhook dispatch code
• Execution timing tracked via time.time() so executionTime reports real milliseconds (delayTime is always 0 in local dev since there's no queue)

Fixes #409
Fixes #410

What was wrong

The three simulation endpoints (_sim_runsync, _sim_stream, _sim_status) in rp_fastapi.py had two issues:

1. Incomplete webhook payload — only the raw handler output ({"output": ...}) was sent to the webhook URL, while the cloud API sends a full envelope with job metadata. Developers building webhook handlers against the local dev server would see breakage on deploy.

2. No webhook on failure — in _sim_runsync and _sim_status, the error check returned early before the webhook code, so failed jobs never triggered a webhook callback. In production, RunPod sends a webhook with "status": "FAILED".

What changed

• Added import time and job_start/execution_time tracking in each method
• Construct full cloud-parity payload before dispatching the webhook
• Moved webhook dispatch above the early error return so it fires on both success and failure
• Updated tests to assert webhook payload structure (all 7 fields) and added a test for error-path webhook delivery

Test plan

• All existing tests pass (pytest tests/test_serverless/test_modules/test_fastapi.py — 8/8)
• New assertions verify webhook payload contains id, status, input, webhook, delayTime, executionTime, output
• New test verifies webhook fires with "status": "FAILED" when handler raises an exception

[Swigler]

Friendly ping — the dev server returns a different webhook shape than cloud, which breaks local testing. This aligns them. Ready for review.