feat(telemetry): emit gen_ai.request.* from tracing plugin (post-#1045)

[planetf1]

Problem

gen_ai.request.temperature, gen_ai.request.top_p, and related request-parameter attributes are not emitted on backend spans today.

A model_options/_REQUEST_PARAM_MAP approach was prototyped in PR #1036 but withdrawn following review from @jakelorocco and @ajbozarth:

• No backend call site passed model_options, so the code was dead.
• Even if wired, the values would be pre-substitution — Mellea backends mutate model_options before handing them to the SDK (defaults merged, provider-specific renames, unsupported keys stripped). The dict visible at the top of start_generate_span is not what goes over the wire.
• Per @jakelorocco: prefer pass-through as gen_ai.model_options.<key> over a hardcoded remap to OTel-standard names, since the naming differs between providers anyway.

Approach

After #1045 lands and BackendTracingPlugin exists:

1. Capture the final resolved model_options dict in the GENERATION_PRE_CALL hook — the point immediately before the SDK call, where values match what the provider receives.
2. Emit each key/value as gen_ai.model_options.<key> (pass-through, not remapped), or decide at that point whether OTel-standard names (gen_ai.request.temperature etc.) are worth the remap given cross-provider naming differences.

Prerequisites

• refactor: move tracing onto plugin/hook pattern #1045 (Phase 1 plugin/hook pattern) must merge first.

Related

Epic #444 Phase 2.
PR #1036 (where the dead-code prototype was withdrawn).