model: replace assert-based validation with raise ValueError (works under python -O)

[SoundMatt]

Problem

The validators in src/vss_tools/model.py use assert COND, "msg"
to enforce ~32 invariants on VSSData, VSSDataDatatype, and
VSSUnit fields. Pydantic's own documentation explicitly warns
against this pattern: Python's -O optimisation flag disables
asserts, allowing invalid data to pass through.

Anyone running vss-tools under python -O (some packaging
pipelines, performance-tuned environments) gets all VSS spec
validation silently bypassed. The parser accepts invalid specs as
valid, and downstream exporters then act on garbage data.

Reproduction

python -O -c "
from vss_tools.model import VSSDataDatatype
# This *should* raise ValidationError ('foo' is not a datatype)
VSSDataDatatype(type='attribute', description='x', datatype='foo')
"
# Before this PR: silently succeeds (wrong).
# After this PR:  raises ValidationError (correct).

Fix

Mechanically replace each:

assert COND, "msg"

with:

if not COND:
    raise ValueError("msg")

Pydantic wraps a ValueError raised inside a validator as a
ValidationError automatically, so externally-observable behaviour
is identical in normal mode. The new form is preserved under
python -O.

Test

tests/test_model.py already parametrises 50+ invalid-input cases
that expect pydantic.ValidationError. All continue to pass after
this PR in normal mode. They would have failed under python -O
before this change; now they pass in both modes.

I considered adding an explicit python -O job to CI to catch
future regressions of this anti-pattern, but kept it out of this PR
to stay focused. Happy to follow up if maintainers want.

Notes

• 32 assert statements replaced across VSSData, VSSUnit, and
  VSSDataDatatype validators.
• A few asserts had not (X and Y) form; those became if X and Y
  for natural inversion. Equivalent semantics, idiomatic Python.
• One pre-existing falsy-check bug (if not self.default: in
  check_default_min_max) is intentionally left untouched — it's
  a separate validation bug that deserves a focused PR (signal
  with default=0, min=10 silently passes today).

[sschleemilch]

Yeah, that's a valid one

[SoundMatt]

CI caught a regression: six tests in test_description_error/ and
test_overlay/ asserted on the literal string 'type': 'assertion_error'
in pydantic's error output, which becomes 'type': 'value_error' after
the assert -> raise refactor. Pushed a follow-up commit updating those
test expectations. Same coverage, same intent — only the type-tag string
differs.

[erikbosch]

MoM:

• Presented at meeting
• ok to merge