Restructure: descriptive tier and experiment names, paper/manuscript

- paper/pnas -> paper/manuscript (venue-neutral)
- configs/layer1 -> configs/inheritance, src/knowledge -> src/inheritance
  (imported as `inheritance`), make layer1 -> make inheritance; layer2 alias dropped
- inheritance and trained-network bundles named after the manuscript figure
  they feed (fig2_grounding_sweep, figS3_rebaselining, ...), or descriptively
  where they feed none; configs keep their `experiment:` value so parquet
  hashes are unchanged, only output.dir moves
- figure scripts, SI figure sources, notebooks, REPRODUCING.md, README and the
  SI Methods/tables updated; make clean no longer deletes tracked manifests;
  reproduce.sh hashes the s{seed}/ layouts too

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y64o8FKP7rCuXzC48pxpMm

tests/test_scientific_validation.py

@@ -9,7 +9,7 @@ check (the simulator reproduces known population-genetics results) and a code ch
 (the implementation is correct). If any test here fails, the science is wrong, not
 just the code -- do not trust any downstream Layer-1 figure until these pass.
 
-The Layer-1 implementation in `src/knowledge/` is DONE, for validation purposes,
+The Layer-1 implementation in `src/inheritance/` is DONE, for validation purposes,
 when the conformance tests in Part 5 pass against the real package.
 
 HOW IT IS STRUCTURED
@@ -98,7 +98,7 @@ def theory_union_coverage_fraction(K_T: int, rho: float, q: float) -> float:
 # =====================================================================================
 #
 # This is the smallest correct implementation of the core Layer-1 dynamics. It exists
-# so the spine tests can run before src/knowledge/ is written, and so the exact
+# so the spine tests can run before src/inheritance/ is written, and so the exact
 # semantics the package must reproduce are unambiguous. The package will do far more
 # (config, logging, regions, selection, re-minting, per-region metrics); it must agree
 # with THIS on the analytic-check subset.
@@ -388,8 +388,8 @@ class TestMetricSanity:
 # from the minimal one built here, adapt the *construction* of cfg, never the tolerance.
 
 try:
-    import knowledge.metrics as knowledge_metrics
-    import knowledge.teachers as knowledge_teachers
+    import inheritance.metrics as knowledge_metrics
+    import inheritance.teachers as knowledge_teachers
     HAVE_KNOWLEDGE = True
 except Exception:  # package not built yet -> conformance layer skips, spine still runs
     knowledge_metrics = knowledge_teachers = None
@@ -440,7 +440,7 @@ class TestPackageDynamicsConform:
 
     def _run_lineage(self, cfg_overrides, seed):
         lineage = pytest.importorskip(
-            "knowledge.lineage", reason="Layer-1 package not implemented yet")
+            "inheritance.lineage", reason="Layer-1 package not implemented yet")
         base = {
             "truth": {"K": 50, "R": 1, "tail": "zipf", "zipf_s": 1.1,
                       "tail_frac": 0.5, "tail_threshold": 1e-3},