genai-openai: multimodal content-part arrays are dropped from captured messages (including the text part)

Describe your environment

• opentelemetry-instrumentation-genai-openai 1.1b0 (also reproduced on current main)
• opentelemetry-util-genai 1.1b0
• openai SDK 2.x, Python 3.12
• Content capture opted in: OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT=SPAN_ONLY

What happened?

When a chat completion request uses the OpenAI content-part array form for multimodal messages, e.g.

messages = [
    {"role": "user", "content": [
        {"type": "text", "text": "What is in this image?"},
        {"type": "image_url", "image_url": {"url": "https://example.com/cat.png"}},
    ]},
]

the captured gen_ai.input.messages records the user message with empty parts — the text part is silently dropped along with the image part:

[{"role":"user","parts":[]}]

What did you expect to see?

At minimum the text part should be preserved. Ideally non-text parts are represented too (there is already a GenericPart type in opentelemetry-util-genai documented for exactly this: provider-specific part types that don't map to the standard semconv part types).

Root cause

_prepare_input_messages in instrumentation/opentelemetry-instrumentation-genai-openai/src/opentelemetry/instrumentation/genai/openai/utils.py only appends a part when _is_text_part(content) is true, and _is_text_part accepts only str or an iterable of str:

def _is_text_part(content: Any) -> bool:
    return isinstance(content, str) or (
        isinstance(content, Iterable)
        and all(isinstance(part, str) for part in content)
    )

A content-part array (list of dicts / SDK part models) fails this check, and there is no fallback branch, so the message ends up with no parts at all.

Additional context

This is a regression in observability coverage for anyone migrating from third-party OpenAI instrumentations, which serialized the full content-part array. I have a fix ready (map text parts to TextPart, preserve other part types as GenericPart) and will open a PR referencing this issue.

Start in instrumentation/opentelemetry-instrumentation-genai-openai/src/opentelemetry/instrumentation/genai/openai/utils.py, focusing on _prepare_input_messages and _is_text_part, then inspect the GenericPart type in opentelemetry-util-genai. Done means content-part arrays no longer produce empty parts: text content is captured and non-text content is represented where supported.