Add provider-neutral Claude/OpenAI benchmarks and fix temperature for reasoning models

- benchmarks/benchmark_claude.py: end-to-end recovery + sampled audit +
  classification, defaults anthropic/claude-sonnet-5 with --provider/--model/
  --base-url overrides, smoke + mock modes, per-dataset checkpoints
- benchmarks/benchmark_claude_classification.py: focused correctness harness
  with the same provider options and offline fake client
- OpenAIClient: omit temperature for gpt-5/gpt-6/o-series reasoning models
  (API rejects it with 400); reasoning effort stays at API default (medium)
- benchmark_buffdata PRICING: add gpt-5.6-sol / gpt-5.6 rates

Author: voidstackloop

benchmarks/benchmark_buffdata.py

@@ -162,6 +162,8 @@
 PRICING = {
     "gemini-3.7-flash": {"input": 0.075, "output": 0.30},
     "claude-sonnet-5": {"input": 3.00, "output": 15.00},
+    "gpt-5.6-sol": {"input": 4.00, "output": 20.00},
+    "gpt-5.6": {"input": 4.00, "output": 20.00},
     "default": {"input": 1.00, "output": 5.00},
 }