feat(telemetry): emit llm.prompt_template.* from tracing plugin (post-#1045)

[planetf1]

Problem

llm.prompt_template.template and llm.prompt_template.variables (OpenInference convention) are not emitted on backend spans today.

A component-level approach was prototyped in PR #1036 but withdrawn following review from @jakelorocco: capturing template state on Instruction and GenerativeStub is the wrong layer — it only covers two component types, captures pre-substitution values, and puts telemetry concerns inside domain objects.

The correct extraction point is the formatter render path (TemplateFormatter._render_representation), which is reached for every component type and holds the actual template and variables used at generation time.

Approach

After #1045 lands and BackendTracingPlugin exists:

1. Capture (template_str, variables_dict) at the point _render_representation renders an action — one place, covers all components.
2. In BackendTracingPlugin, emit:
   • llm.prompt_template.template — always (template text is not user content)
   • llm.prompt_template.variables — only when is_content_tracing_enabled() returns true (variables may contain user data)
3. is_content_tracing_enabled() and the MELLEA_TRACE_CONTENT / OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT flag are already wired in mellea/telemetry/tracing.py from feat(telemetry): close five OTel GenAI semantic convention emission gaps (#1035) #1036.

Prerequisites

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

Related

Epic #444 Phase 2 — alongside the existing #1035 entry.
PR #1036 (where the component-level prototype was withdrawn).