fix(core): report output_dim, not min_output_dim, when it is too small

[ChrisW09]

Fixes #38.

Problem

AdaptiveResolutionMixin._resolve_output_bounds reported every floor violation against
min_output_dim, but lo is assigned output_dim on the non-adaptive branch a few lines
earlier. So a user who set output_dim got told about a parameter they never touched — and
one that is ignored entirely when adaptive=False:

Preprocessor(numerical_method="ple", output_dim=0)
-> min_output_dim must be >= 1, got 0.
   Fix: raise min_output_dim to at least the family minimum.

Acting on that "Fix:" line would not have helped.

The two halves of the package also disagreed: families that pre-validate output_dim
themselves reported it correctly, while those relying on the shared resolver did not.

ple          output_dim=0 -> min_output_dim must be >= 1, got 0.
cubicspline  output_dim=0 -> output_dim must be >= 3 for the cubic spline basis, got 0

Fix

Pick the name from the branch that produced the value — min_output_dim only when
adaptive is set and the caller supplied one, otherwise output_dim. The "Fix:" line
follows suit.

ple                        output_dim=0     -> output_dim must be >= 1, got 0.
cubicspline                output_dim=0     -> output_dim must be >= 3 ...
adaptive, min_output_dim=0                  -> min_output_dim must be >= 1, got 0.

Tests

Three added to tests/test_exceptions.py: the non-adaptive message names output_dim, it
does not mention min_output_dim, and the adaptive path still names min_output_dim when
that is genuinely what was set.

Cosmetic, but it is the first thing a user sees when they mis-size a transformer, and it was
pointing at the wrong knob.

Full suite: 483 passed, 9 xfailed. ruff check clean on changed files; pyright unchanged at
its pre-existing 71 errors.

🤖 Generated with Claude Code