Add template cProfile runner

[ReguiguiMohamed]

Summary

• add a small runner for profiling any template with Python's built-in cProfile
• write a binary profile and cumulative-time report
• document workload requirements without changing normal template execution

Why

The templates have different dependencies and workloads. Running cProfile in
a child process keeps profiler code out of each example and leaves normal
template execution unchanged.

Generated traces remain local because profiler output depends on the workload, dependency versions, and hardware.

Related to pyfenn/fenn#146.

Validation

• ruff check --isolated profiling/profile_template.py
• ruff format --check --isolated profiling/profile_template.py
• python -m compileall -q profiling/profile_template.py
• smoke-tested against the empty template
• verified unknown template names fail before execution

[ApusBerliozi]

Hi @ReguiguiMohamed, thx for your work man!

Could you please provide us with an example of profiling result for mlp-binary? Screenshot will do just fine

[ReguiguiMohamed]

Sure, here is an example cProfile result generated for mlp-binary.

Environment: Python 3.12.4, PyTorch 2.12.0+cpu, running on CPU.