Drop autonomous-vs-full size invariant from test_full_is_largest

The `llms-full.txt > llms-autonomous.txt` length assertion was a
stale Wave-1 invariant. Original assumption: "full" is the
comprehensive reference and therefore the longest by definition.
Reality after Wave 1 + Wave 2 + 17 review rounds: the autonomous
reference guide has grown organically (estimator-support matrix,
per-design-feature reasoning, three worked examples, contract
citations like `had.py:1450-1459`) while `llms-full.txt` is the
API-docs roll-up and hasn't been updated in lockstep with
Wave 2's agent-facing content. The two guides serve different
audiences — there is no user-facing invariant requiring one to be
larger than the other.

The invariant was forcing artificial trims: each review round
that added a contract caveat to autonomous required compensating
prose deletion elsewhere just to keep autonomous below full. Each
trim made autonomous marginally less complete without
methodological benefit.

Removed the autonomous comparison; kept `full > concise` and
`full > practitioner` since `concise` is by definition shorter and
`practitioner` is workflow prose without the API roll-up. Added a
test docstring explaining the rationale so future editors don't
reintroduce the comparison.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: igerber

tests/test_guides.py

@@ -20,10 +20,19 @@ def test_default_is_concise():
 
 
 def test_full_is_largest():
-    lengths = {v: len(get_llm_guide(v)) for v in ("concise", "full", "practitioner", "autonomous")}
+    """`llms-full.txt` is the API-docs roll-up; it should remain larger
+    than the short `concise` summary and the workflow-prose
+    `practitioner` guide. The `autonomous` reference guide is
+    deliberately excluded from this comparison: it serves a different
+    audience (LLM agents reasoning about estimator choice) and has
+    grown organically through Wave 1 + Wave 2 review rounds with
+    estimator-matrix detail, worked examples, and contract citations
+    that don't have a counterpart in `llms-full.txt`'s API roll-up.
+    Either of the two can be larger without violating any user-facing
+    invariant."""
+    lengths = {v: len(get_llm_guide(v)) for v in ("concise", "full", "practitioner")}
     assert lengths["full"] > lengths["concise"]
     assert lengths["full"] > lengths["practitioner"]
-    assert lengths["full"] > lengths["autonomous"]
 
 
 def test_content_stability_practitioner_workflow():