New package: unhandled Exceptions logging instrumentation

[iblancasa]

This ticket is to add a new contrib instrumentation package that captures uncaught exceptions and unhandled asyncio task exceptions and emits them as OpenTelemetry logs. This provides a consistent, out‑of‑the‑box error tracking signal across services without custom handlers, and ensures "last error" telemetry reaches the standard logs pipeline.

Backend services frequently fail without producing telemetry:

• Uncaught exceptions terminate processes without a log record.
• Unhandled asyncio task exceptions are only printed to stderr.

Teams lose stack traces, severity, and a consistent error tracking experience.

Once one of those events happens, the new instrumentation can capture the Exception and create a log message following the OpenTelemetry semantic conventions for exception: https://opentelemetry.io/docs/specs/semconv/exceptions/exceptions-logs/

[iblancasa]

I will draft a PR with the new package.

[dgenio]

I’m interested in implementing this contrib package.

Initial design for PR1:

• Hook sys.excepthook (and threading.excepthook where available), chaining any existing hooks.
• For asyncio: install a loop exception handler to capture unhandled task exceptions (“Task exception was never retrieved”).
• Emit OTEL logs following exception semantic conventions (type/message/stacktrace + severity mapping).
• Provide minimal config flags (enable/disable each hook, include stacktrace).
• Tests: subprocess test for truly uncaught exceptions + asyncio unit tests for unhandled task exceptions.

Two questions before I start:

1. Minimum supported Python version for this instrumentation?
2. Preferred package name/layout (e.g., opentelemetry-instrumentation-unhandled-exceptions) following the standard contrib structure?

If that matches expectations, I’ll start with the core hooks + tests and iterate from there.