[grpc] AIO server spans record uncaught exceptions twice

Describe your environment

• Python: 3.12.13
• grpcio: 1.78.1
• opentelemetry-api/sdk: 1.44.0
• opentelemetry-instrumentation-grpc: 0.65b0
• gRPC API: grpc.aio server

The current main implementation appears to have the same behavior.

What happened?

When a unary AIO gRPC handler raises an uncaught exception subclass (for example, ValueError or aiohttp.ClientConnectorError), the server span contains two identical exception events.

The exception is recorded once explicitly in _intercept_aio_server_unary:

except Exception as error:
    if type(error) != Exception:
        span.record_exception(error)
    raise error

It is then recorded again when the exception leaves the context returned by _start_span(). _start_span() calls start_as_current_span() with set_status_on_exception=False, but leaves record_exception=True (the default).

The same pattern exists in _intercept_aio_server_stream.

Steps to reproduce

1. Instrument a grpc.aio server with GrpcAioInstrumentorServer.
2. Configure an SDK tracer provider and an in-memory or OTLP exporter.
3. Invoke a unary handler that raises an exception subclass.
4. Inspect the exported server span events.

Observed result:

span.events = [exception, exception]

Both events contain the same exception type, message, and stack trace. One stack trace originates at the explicit span.record_exception(error) call; the other originates from opentelemetry.trace.use_span() when the re-raised exception exits start_as_current_span().

Expected behavior

A single uncaught exception should produce one exception event on the gRPC server span.

Suggested fix

Pass record_exception=False when creating the server span because the interceptor already records qualifying exceptions explicitly:

self._tracer.start_as_current_span(
    ...,
    record_exception=False,
    set_status_on_exception=False,
)

Alternatively, remove the explicit span.record_exception(error) and rely on the context manager, provided the existing special handling for bare Exception/gRPC aborts is preserved.

Additional context

Relevant source:

• AIO interceptor: https://github.com/open-telemetry/opentelemetry-python-contrib/blob/main/instrumentation/opentelemetry-instrumentation-grpc/src/opentelemetry/instrumentation/grpc/_aio_server.py
• Server span creation: https://github.com/open-telemetry/opentelemetry-python-contrib/blob/main/instrumentation/opentelemetry-instrumentation-grpc/src/opentelemetry/instrumentation/grpc/_server.py

Start with the AIO interceptor in instrumentation/opentelemetry-instrumentation-grpc/src/opentelemetry/instrumentation/grpc/_aio_server.py and server span creation in _server.py. Trace unary and stream exception handling, then reproduce an uncaught exception and verify the exported span contains one exception event while preserving the existing bare Exception and gRPC abort handling.