fix: floor token budgets and surface truncation, add live progress

Reasoning models burn thousands of completion tokens before the
answer. Small per-test caps caused silent truncation and zero scores.
The runner now floors max_tokens at the config default (12000). The
client parses finish_reason. Reports mark truncated cells and the
JSON format carries response text for debugging. A stderr progress
reporter prints one line per completed request; --quiet disables it.

config.example.yaml

@@ -4,6 +4,8 @@ models:
   - uni/deepseek-v4-flash-0731
   - uni/qwen3.6-27b
   - uni/btl-4
-concurrency: 4
-request_timeout: 120s
-max_tokens_default: 4000
+concurrency: 8         # default when omitted; matches internal/config's default
+request_timeout: 300s  # default when omitted
+max_tokens_default: 12000  # default when omitted; a reasoning model can spend
+                            # thousands of tokens on reasoning_content before
+                            # writing an answer, so this must stay generous