🐛 Bug Report: NotGiven sentinel leak in Anthropic instrumentation

Which component is this bug for?

Anthropic Instrumentation

📜 Description

set_span_attribute() in \traceloop\openllmetry\tree\main\packages\opentelemetry-instrumentation-anthropic\opentelemetry\instrumentation\anthropic\utils.py filters None and empty strings but not the Anthropic SDK's NOT_GIVEN sentinel (type Omit / NotGiven). When optional parameters like temperature or top_p are not explicitly set, the sentinel leaks into span.set_attribute(), producing OTel type warnings:

Invalid type Omit for attribute 'gen_ai.request.temperature' value.
Expected one of ['bool', 'str', 'bytes', 'int', 'float'] or a sequence of those types

Same pattern as #4230(log Body) and #3472(OpenAI Omit), but in Anthropic span attributes.

👟 Reproduction steps

1. Install opentelemetry-instrumentation-anthropic and anthropic SDK (0.52+)
2. Call client.messages.create(model="...", max_tokens=1024, messages=[...]) without setting temperature or top_p
3. Observe warning: Invalid type Omit for attribute 'gen_ai.request.temperature'

👍 Expected behavior

set_span_attribute() should silently skip NOT_GIVEN / Omit sentinel values, same as it skips None and "".

👎 Actual Behavior with Screenshots

The sentinel passes through to span.set_attribute(), which rejects non-primitive types and logs a warning per attribute.

🤖 Python Version

3.11 / 3.13 (reproducible on both)

📃 Provide any additional context for the Bug.

The OpenAI instrumentation already handles this in two places:
・_set_span_attribute checks value == openai.NOT_GIVEN
・_sanitize_sentinel_values() strips sentinels from kwargs(#3472)

👀 Have you spent some time to check if this bug has been raised before?

• I checked and didn't find similar issue

Are you willing to submit PR?

Yes I am willing to submit a PR!

Start in traceloop/openllmetry/tree/main/packages/opentelemetry-instrumentation-anthropic/opentelemetry/instrumentation/anthropic/utils.py and inspect set_span_attribute(), comparing its filtering with the OpenAI instrumentation references described in the issue. Reproduce the messages.create() call without temperature or top_p, then verify that Anthropic NOT_GIVEN/Omit values are skipped and no invalid OTel attribute warnings are emitted.