Add advection velocity to species mapping + 'user defined' parameter role

[jcschaff]

Cherry-picked from @LDre398 (Logan Drescher), then reviewed, fixed (rebase gaps), tested and linted.

What's in here

Advection velocity on species mappings — round-trips a <Velocity X= Y= Z=> element under LocalizedCompoundSpec:

• SpeciesMapping gains velocity_x/_y/_z (float | str | None), included in expressions for field-data extraction;
• reader visit_Velocity parses the element; writer emits it, writing only non-"0.0" components.

Parameter role — Model.add_model_parameter(...) now defaults role="user defined" (and accepts an override), matching VCell's Java representation (was "model_parameter").

Reader method rename — generic_visit → generic_visit_children (clearer intent), updated across the visitor.

New SASCO fixtures + tests exercise the feature end-to-end (load → write → simulate → plot), including a velocity model that yields 21 velocity output channels.

Review notes — rebase gaps I fixed

These cherry-picks were authored before the data-model split (#41) and the lenient-reader work (#44), so the rebase left three breakages:

1. Dropped velocity fields. SpeciesMapping moved from models.py to models_app.py, so the rebase lost the velocity_x/_y/_z field additions. The reader's node.velocity_x = … would raise on a pydantic model with no such field — i.e. any VCML with a <Velocity> element failed to load. Restored the fields.
2. Incomplete rename. generic_visit → generic_visit_children missed visit_Parameter (vcml_reader.py:212); the stale call was silently swallowed by the best-effort backstop (a caught AttributeError per <Parameter>, plus dropped nested-parameter children). Fixed the call site.
3. Unregistered fixtures. The two SASCO fixtures were added to model_fixtures.py but not imported in conftest.py, so the new tests errored with "fixture not found". Registered them.

Also tidied test_results_writing.py (removed debug prints / commented-out dead code, added a channel_data assertion, run cleanup() in finally).

Verification

• Velocity parses and round-trips through reader→writer (smoke-tested); existing reader/writer round-trip equality is preserved (those fixtures have no velocity → no regression).
• make check green — ruff, mypy (312 files), deptry, uv lock --check.
• New SASCO simulation tests pass (incl. len(velocity_channels) == 21); full tests/vcml suite → 43 passed, 5 skipped.

Note

The SASCO tests run real FV simulations (small 16×31×1 2D mesh, EndTime=10 — ~50s total), so they add some CI time but validate the velocity output end-to-end.

🤖 Generated with Claude Code