Add agent rehearsal harness, findings, and per-run evidence for split-loop-with-lambda

Six scripted pi sub-agent runs (3 tool regimes × 2 models) rehearsed the
exercise; deterministic gate = solution test + approved file. Key results:
mandated CLI triples cost on the strong model (4:06 vs 0:39, PASS both);
weak model fails only under the mandate (A+Q), passes when free or optional.
Drift documented with code: scorer.updateSelection left inside the extracted
method (A+Q, C+Q). Harness in tools/rehearse; summary in
rehearsal-findings.md; per-run evidence in rehearsal-runs/; cheat sheet in
rehearse-with-agent.md corrected (lambda-typed parameters unsupported).

ideas/split-loop-with-lambda/exercise/rehearsal-runs/C/measure.txt

@@ -0,0 +1,35 @@
+--- Run C ---
+session file: /tmp/rehearsal/runs/C/session/2026-09-27T15-45-09-782Z_01a0e38a-a696-7513-8cb3-d7e75370aeff.jsonl
+model:                z-ai/glm-5.3-flash
+session span:         156.5 s
+assistant messages:   20
+tool calls:           23 total
+  bash                     15
+  read                     4
+  edit                     4
+tokens input:         13250
+tokens output:        8393
+tokens cacheRead:     124544
+tokens cacheWrite:    0
+tokens reasoning:     4618
+tokens TOTAL (sum):   150805
+--- /usr/bin/time -v (wall clock + max RSS) ---
+	Elapsed (wall clock) time (h:mm:ss or m:ss): 2:36.61
+	Maximum resident set size (kbytes): 248952
+	Exit status: 0
+--- pi stdout (last 15 lines) ---
++        }
++        return expectedWeather;
++    }
+ }
+```
+
+Supporting change in `src/test/java/codingdojo/LongRangeForecastsTest.java`:
+
+```diff
+-        var quarter = "Q3";
+-        var actual = new LongRangeForecasts().longRangeForecast(quarter);
++        var actual = new LongRangeForecasts().longRangeForecast();
+```
+
+Stopping here as instructed.