opentelemetry-sdk: stop coercing instrumentation options once instrumentors read ConfigProvider

Is your feature request related to a problem?

configure_instrumentation() in opentelemetry-sdk/src/opentelemetry/sdk/_configuration/instrumentation.py, added in https://github.com/open-telemetry/opentelemetry-python/pull/5372, does two separate jobs:

1. Activation. Load the opentelemetry_instrumentor entry point by name, honor enabled: false, skip an already instrumented library, and isolate a failing instrumentor.
2. Option delivery. Read the instrumentor's configuration dataclass attribute, run the raw options through _dict_to_dataclass, and forward the result as keyword arguments to instrument().

Job 2 belongs on the instrumentation side. The Instrumentation Configuration API specification states that instrumentation libraries access ConfigProvider during initialization, either because it is passed to them or because they read it from a global. See https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/configuration/api.md.

Raised by @xrmx in https://github.com/open-telemetry/opentelemetry-python-contrib/pull/4766#issuecomment-5356515506.

Leaving job 2 in place after instrumentors can read configuration for themselves means two components coerce the same values, and the SDK keeps instrumentation specific knowledge it does not need.

Describe the solution you'd like

Once BaseInstrumentor reads its own configuration from ConfigProvider, remove the coercion branch from configure_instrumentation():

• Drop the getattr(cls, "configuration", None) lookup, the is_dataclass check, and the _dict_to_dataclass call.
• Drop the rebuild of options from the dataclass fields.
• Keep activation unchanged: entry point loading, enabled: false, the already instrumented check, and the error isolation.
• Keep passing the remaining raw options to instrument(), or decide explicitly that all options now arrive through ConfigProvider and the call becomes instrument() with no arguments.

Update docs/sdk/configuration.rst to describe where instrumentation options are read.

Describe alternatives you've considered

Leave the branch in place. It becomes unreachable once instrumentation reads its own configuration, so it is dead code that still has to be maintained and tested.

Keep both paths permanently and define a precedence between them. This doubles the number of places a user has to look at to understand where an option came from, for no benefit.

Additional Context

Ordering, because this work spans two repositories:

• Depends on ConfigProvider and ConfigProperties from https://github.com/open-telemetry/opentelemetry-python/pull/5486.
• Depends on the BaseInstrumentor read hook tracked in https://github.com/open-telemetry/opentelemetry-python-contrib/issues/5027.
• Must land after that contrib work, not before.

If this removal shipped first, options set in a configuration file would be silently dropped for every instrumentor, because nothing would read them.

opentelemetry-sdk/tests/_configuration/test_instrumentation.py covers the coercion behavior today. The tests for configuration coercion, None field omission, and unknown field rejection move to the contrib repository along with the behavior. The activation tests stay here.

Would you like to implement a fix?

Yes

Start with opentelemetry-sdk/src/opentelemetry/sdk/_configuration/instrumentation.py and its activation tests in opentelemetry-sdk/tests/_configuration/test_instrumentation.py. Check the ConfigProvider and BaseInstrumentor dependencies in the linked work before changing behavior. Done means activation remains covered, instrumentation options are read through ConfigProvider, docs/sdk/configuration.rst is updated, and coercion tests move to contrib.