Extract Vis._identity_args() helper to unify delegate call signatures

[gerchowl]

Tier 2 — small refactor

Three delegation sugar properties use three different call shapes for
the same identity triple, flagged by the adversarial review
(`src/pymat/vis/_model.py`):

# .mtlx — kwarg-style
return self.client.mtlx(self.source, self.material_id, tier=self.tier)

# .channels — positional
return self.client.channels(self.source, self.material_id, self.tier)

# .materialize — positional
return self.client.materialize(self.source, self.material_id, self.tier, output_dir)

If mat-vis-client ever makes `tier` keyword-only (`*, tier: str = "1k"`),
`.channels` + `.materialize` break while `.mtlx` keeps working — and the
breakage is silent-ish (`TypeError: got positional arg for keyword-only`).

Fix

def _identity_args(self) -> tuple[str, str, str]:
    """Return (source, material_id, tier) — the positional arg triple
    that mat-vis-client methods expect. Call guarded by has_mapping."""
    return (self.source, self.material_id, self.tier)

@property
def mtlx(self) -> MtlxSource | None:
    if not self.has_mapping:
        return None
    src, mid, tier = self._identity_args()
    return self.client.mtlx(src, mid, tier=tier)

# Same shape for .channels and .materialize.

Test

def test_identity_args_tuple(self):
    v = Vis(source="a", material_id="b", tier="2k")
    assert v._identity_args() == ("a", "b", "2k")

Plus: grep for every remaining `self.client.X(self.source, self.material_id, ...)`
and route through the helper. This is a consistency refactor — not
behavioral.

Why a helper instead of explicit calls everywhere

The helper makes it one place to update when mat-vis-client changes
signatures. Also provides a seam for testing (monkeypatch
`_identity_args` to return a fixture triple).