[BUG] Calculator availability check uses schema importability instead of engine availability, causing runtime failures

[tdpham2]

Problem: When fairchem-core (or mace-torch) is not installed, ChemGraph still selects FAIRChem as the default calculator and advertises it as available. The workflow then fails at runtime with:
ERROR:chemgraph.agent.llm_agent:Error running workflow single_agent: fairchem is not installed.

The default calculator selection in src/chemgraph/schemas/ase_input.py:51-57 checks whether the Pydantic schema class (FAIRChemCalc) is importable:

if FAIRChemCalc:
    default_calculator = FAIRChemCalc
elif MaceCalc:
    default_calculator = MaceCalc
else:
    default_calculator = NWChemCalc

But FAIRChemCalc is always importable because it's a Pydantic BaseModel defined in ChemGraph(src/chemgraph/schemas/calculators/fairchem_calc.py:15). The only top-level dependency it needs is torch, which is a core dependency. The actual engine import (from fairchem.core import ...) is guarded by try/except inside fairchem_calc.py, so the schema class always exists regardless of whether fairchem-core is installed.

Same issue with MaceCalc.

Related problem: While config.toml defines [chemistry.calculators] default = "mace_mp", this value is never read by the calculator selection logic. It's only used for Streamlit UI display.