correctness fixes from gpt-5.5 review (ordinal path + fail-fast asserts)

Second external review (gpt-5.5, correctness-focused) on the post-cleanup tree.
Found no off-by-one/double-flip in the ordinal canonicalization+keying. Fixed the
parts I agreed with and could verify on the path the experiment uses:
- read.py: NaN-safe answer-token renorm. p_a/pmass poisons the profile with NaN when
  pmass underflows to 0 at coherence collapse -- exactly when pmass should just flag it.
  softmax(logp_allowed) is identical when pmass>0 and stable at collapse.
- maps.ipsative_pca: move SVD sign-stabilization INTO the helper so it and
  plot_ipsative_pca share one orientation (saved coords could otherwise mirror the figure).
- instrument: assert ordinal answer_space is ['1'..scale_max] IN ORDER (reduce_ordinal
  weights by position; a reordered space silently inverts E) -- was length-only.
- instrument.per_item_categorical: assert per-item dimension/sign agree across frames and
  frames are distinct, instead of silently averaging under rows[0]'s metadata.
- pyproject: move matplotlib+textalloc to an optional `maps` extra; evals stay headless.
- tests: drop imports of the deleted reduce_nominal/expected_value, inline the expectation,
  remove the now-impossible nominal-reducer test.

Deferred (flagged to maintainer): two NaN/window issues in guided.py's forced-choice
rollout (nominal evaluate() path) -- not exercised by this experiment, can't smoke-test,
and the NaN-as-collapse-signal there is a deliberate design.

Verified: experiment smoke green on all 4 instruments (no assert false-fires), 6 pure
unit tests pass, headless import clean, 16pf map renders.

Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>

tests/test_instrument.py

@@ -8,13 +8,17 @@ import numpy as np
 
 from tinymfv.instrument import (
     Instrument, InstrItem, canonicalize_to_forward, per_item_categorical,
-    reduce_nominal, reduce_ordinal, expected_value, shuffle_dimensions,
+    reduce_ordinal, shuffle_dimensions,
 )
 
 M = 5
 ONEHOT = {d: np.eye(M)[d - 1] for d in range(1, M + 1)}  # ONEHOT[4] = mass on scale point 4
 
 
+def _E(p, scale_max):  # expected scale point; reduce_ordinal computes this inline now
+    return float((np.asarray(p) * np.arange(1, scale_max + 1)).sum())
+
+
 def _ord_instr(items):
     return Instrument("t", "endorsement", "ordinal", ["1", "2", "3", "4", "5"],
                       ["care", "authority"], items, prefill="(")
@@ -33,7 +37,7 @@ def test_forward_expectation():
     rows = [{"id": "1", "frame": "forward", "p": ONEHOT[4], "pmass_allowed": 1.0,
              "dimension": "care", "sign": 1, "human_label": None}]
     items = per_item_categorical(rows, "ordinal")
-    assert abs(expected_value(items["1"]["p"], M) - 4.0) < 1e-9
+    assert abs(_E(items["1"]["p"], M) - 4.0) < 1e-9
     prof = reduce_ordinal(items, _ord_instr([]))
     assert abs(prof[0] - 4.0) < 1e-9  # care
     assert np.isnan(prof[1])          # authority has no items
@@ -45,8 +49,8 @@ def test_frame_consistency():
             "dimension": "care", "sign": 1, "human_label": None}]
     inv = [{"id": "1", "frame": "inverted", "p": ONEHOT[5], "pmass_allowed": 1.0,
             "dimension": "care", "sign": 1, "human_label": None}]
-    e_fwd = expected_value(per_item_categorical(fwd, "ordinal")["1"]["p"], M)
-    e_inv = expected_value(per_item_categorical(inv, "ordinal")["1"]["p"], M)
+    e_fwd = _E(per_item_categorical(fwd, "ordinal")["1"]["p"], M)
+    e_inv = _E(per_item_categorical(inv, "ordinal")["1"]["p"], M)
     assert abs(e_fwd - 1.0) < 1e-9 and abs(e_inv - 1.0) < 1e-9  # canonicalization makes them agree
 
 
@@ -59,7 +63,7 @@ def test_keying_is_not_double_flip():
     rows = [{"id": "x", "frame": "inverted", "p": ONEHOT[5], "pmass_allowed": 1.0,
              "dimension": "care", "sign": -1, "human_label": None}]
     items = per_item_categorical(rows, "ordinal")
-    assert abs(expected_value(items["x"]["p"], M) - 1.0) < 1e-9      # canonical agreement-with-item = 1
+    assert abs(_E(items["x"]["p"], M) - 1.0) < 1e-9      # canonical agreement-with-item = 1
     prof = reduce_ordinal(items, _ord_instr([]))
     assert abs(prof[0] - 5.0) < 1e-9                                 # keyed factor score = 5, not 1 (no double flip)
 
@@ -81,15 +85,6 @@ def test_frame_spread_diagnostic():
     assert abs(items["1"]["frame_spread"] - 2.0) < 1e-9
 
 
-def test_reduce_nominal():
-    instr = Instrument("mfv", "salience", "nominal", ["care", "authority"],
-                       ["care", "authority"], [InstrItem("1", "v1")], prefill="(")
-    rows = [{"id": "1", "frame": "forward", "p": np.array([0.8, 0.2]), "pmass_allowed": 1.0},
-            {"id": "2", "frame": "forward", "p": np.array([0.4, 0.6]), "pmass_allowed": 1.0}]
-    prof = reduce_nominal(per_item_categorical(rows, "nominal"), instr)
-    assert np.allclose(prof, [0.6, 0.4])  # mean choice frequency
-
-
 def test_negative_control_shuffle():
     rng = np.random.default_rng(0)
     rows = [{"id": str(i), "frame": "forward", "p": ONEHOT[(i % 5) + 1], "pmass_allowed": 1.0,