fix(grpcio): properly record and re-raise exceptions

[CagriYonca]

Problem Description

Fixes #895

src/instana/instrumentation/grpcio.py::create_span caught gRPC exceptions and recorded them on the span, but did not re-raise them. As a result, the wrapper implicitly returned None, causing:

• TypeError: cannot unpack non-iterable NoneType object when using gRPC interceptors / with_call (e.g. pymilvus).
• Silently swallowed grpc.RpcError exceptions on standard calls.

Solution

• Re-raise wrapped exceptions: Added raise in create_span and call_behavior_with_instana after span.record_exception(exc) so caller applications receive the original grpc.RpcError.
• Isolate telemetry errors: Wrapped Instana internal metadata injection and attribute collection in a dedicated try/except block (logger.debug) to ensure sensor errors never affect customer code.
• Prevent duplicate error count: Configured record_exception=False on start_as_current_span to avoid duplicate ec increments on raise.
• Tests: Added test_server_error_with_call in tests/frameworks/test_grpcio.py to cover with_call exception propagation.