Implement Julia precompilation to reduce first-use latency

[anvika-singhal]

Reduce first-use latency in komamripy by precompiling common KomaMRI workflows using PrecompileTools.jl. First import now triggers precompilation. Subsequent imports are instant due to cached bytecode.

Implemented

src/komamripy/precompile.jl (new)

• Uses @setup_workload and @compile_workload from PrecompileTools
• Precompiles common entry points: Scanner(), brain_phantom2D(), PulseDesigner.EPI_example(), default_sim_params()
• Gracefully skips if PrecompileTools unavailable (wrapped in try/catch)
• Silently redirects stderr to avoid noise during precompilation

src/komamripy/_session.py (modified)

• Added _run_precompilation(jl) function called after using KomaMRI
• Prints "Precompiling KomaMRI..." on first import
• Handles Windows path separators correctly (converts \ to / for Julia)
• Non-fatal error handling: precompilation failures don't block Julia startup

Details

First import (with precompilation):

1. User imports komamripy
2. Julia initializes and loads KomaMRI
3. Precompile script runs, compiling common workflows
4. Bytecode cached in ~/.julia/compiled/
5. ~seconds latency (one-time cost)

Subsequent imports (cached):

1. User imports komamripy
2. Julia loads cached bytecode
3. ~milliseconds latency (instant)

Backwards Compatibility

• No API changes
• PrecompileTools already a dependency of KomaMRI.jl
• Precompilation is optional. If it fails or is unavailable, komamripy still works normally
• All existing code continues to work without modification

Testing

Verified locally:

• First import shows "Precompiling KomaMRI..." message
• Subsequent imports skip precompilation (no message)
• Julia resolves correctly on Windows (path separators handled)
• Precompilation failures don't block startup