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.

internal/report/markdown.go

@@ -20,14 +20,20 @@ func renderMarkdown(w io.Writer, tests []testkit.Test, models []string, results
 	ew.println()
 	scoreHeader := append([]string{"Test", "Category", "Subcategory"}, models...)
 	writeMarkdownHeader(ew, scoreHeader)
+	anyTruncated := false
 	for _, t := range sortedTests(tests) {
 		row := []string{t.ID, t.Category, t.Subcategory}
 		for _, model := range models {
 			r, ok := idx[t.ID][model]
 			row = append(row, cellText(r, ok))
+			anyTruncated = anyTruncated || r.Truncated()
 		}
 		writeMarkdownRow(ew, row)
 	}
+	if anyTruncated {
+		ew.println()
+		ew.println(truncatedLegend)
+	}
 
 	ew.println()
 	ew.println("## Category rollup (mean score)")