Instrument the sampling loop to track each iteration attempt, providing visibility into retry behavior

[psschwei]

Description:
Instrument the sampling loop to track each iteration attempt, providing visibility into retry behavior.

Detailed Requirements:

1. Add spans for sampling operations:
   • sampler.sample - Overall sampling operation
   • sampler.iterate - Single sampling iteration
   • sampler.repair - Repair attempt
   • sampler.select - Selection from candidates
2. Add attributes:
   • strategy - Sampling strategy name
   • iteration - Current iteration number
   • max_iterations - Maximum allowed iterations
   • status - success/failure/repair
3. Track iteration outcomes:
   • Add span events for validation pass/fail
   • Record reason for rejection

Files to Modify:

• mellea/stdlib/sampling/base.py - Instrument BaseSamplingStrategy
• mellea/stdlib/sampling/rejection.py - Instrument RejectionSamplingStrategy
• mellea/stdlib/sampling/sofai.py - Instrument SoFAISamplingStrategy
• mellea/stdlib/sampling/majority_voting.py - Instrument MajorityVotingStrategy

Span Hierarchy Example:

sampler.sample
├── sampler.iterate (attempt 1)
│   ├── backend.request
│   └── validator.check (FAIL)
├── sampler.iterate (attempt 2)
│   ├── backend.request
│   └── validator.check (PASS)
└── [success]

Acceptance Criteria:

• Each sampling iteration has its own span
• Iteration count and outcomes tracked
• Strategy-specific behavior captured
• Spans correctly nested

[ajbozarth]

Closing as part of the tracing epic (#444) re-plan. This sub-issue predates the plugin/hooks pattern (PR #653) and the GenerationMetadata partition (PR #908); its spec has drifted.

Replacement: part of #1050 — sampling, validator, and formatter spans via plugin hooks (rolled together since the mechanism is identical across all three).

See #444 for the full refreshed plan.