Add support for emitting inference events and enrich message types

[Cirilla-zmh]

Description

• Add support for emitting inference events.
• Add gen_ai.system_instructions attributes for inference span or events.
• Refactor span_utils.py for more brief implementation.
• Add Reasoning, Blob, File, Uri message types.

Fixes #3065

Type of change

Please delete options that are not relevant.

• New feature (non-breaking change which adds functionality)
• This change requires a documentation update

How Has This Been Tested?

Please describe the tests that you ran to verify your changes. Provide instructions so we can reproduce. Please also list any relevant details for your test configuration

• Add unit tests

Does This PR Require a Core Repo Change?

• Yes. - Link to PR:
• No.

Checklist:

See contributing.md for styleguide, changelog guidelines, and more.

• Followed the style guidelines of this project
• Changelogs have been updated
• Unit tests have been added
• Documentation has been updated

[Cirilla-zmh]

Typecheck failure is not related to this PR, so it may be ignored. cc @aabmass

typecheck: commands[0]> pyright
/home/runner/work/opentelemetry-python-contrib/opentelemetry-python-contrib/util/opentelemetry-util-genai/src/opentelemetry/util/genai/_upload/completion_hook.py
  /home/runner/work/opentelemetry-python-contrib/opentelemetry-python-contrib/util/opentelemetry-util-genai/src/opentelemetry/util/genai/_upload/completion_hook.py:175:17 - error: No overloads for "join" match the provided arguments (reportCallIssue)
  /home/runner/work/opentelemetry-python-contrib/opentelemetry-python-contrib/util/opentelemetry-util-genai/src/opentelemetry/util/genai/_upload/completion_hook.py:175:27 - error: Argument of type "Generator[str | Unknown | bytes | Any, None, None]" cannot be assigned to parameter "iterable" of type "Iterable[str]" in function "join"
    "Generator[str | Unknown | bytes | Any, None, None]" is not assignable to "Iterable[str]"
      Type parameter "_T_co@Iterable" is covariant, but "str | Unknown | bytes | Any" is not a subtype of "str"
        Type "str | Unknown | bytes | Any" is not assignable to type "str"
          "bytes" is not assignable to "str" (reportArgumentType)

[JWinermaSplunk]

I think changing some of the apply attributes methods to output a dictionary instead of attaching the attributes for events sake is ok, though making the methods llm specific will lead to a lot of overlapping code overtime if more genai types are implemented in the future. More so applies to methods like _apply_common_span_attributes, _apply_finish_attributes, etc, which could take different genai type invocations.

[Cirilla-zmh]

More so applies to methods like _apply_common_span_attributes, _apply_finish_attributes, etc, which could take different genai type invocations.

@JWinermaSplunk Thanks for your suggestions! We do need to classify the attributes for common, inference, agent and so on to reuse them as much as possible. However, I don't want to involve many features in this PR because that means we'd better refactor the design of LLMInovcation to a layered model with CommonInvocation, MessagesInocation and so on.

Could you please create another issue to mark this proposal? I believe @keith-decker and @aabmass would also be interested in this.