Optimise generic sum-factorisation lowering

[pbrubeck]

Summary

• choose contraction plans from bounded, pruned factorisation candidates
• lower ragged contractions to exact iteration domains
• preserve shared physically mapped tabulations as finite-element structure
• add reproducible Johnson--Mercier code-generation metrics

Performance

Generated Johnson--Mercier metrics with the paired FIAT branch:

dim	compile	FLOPs	scalar temps	array temps	elements	AST lines
2D	0.507 s	39,619	41	29	1,509	196
3D	3.211 s	1,016,342	130	69	17,992	530

Validation

• TSFC sum-factorisation and refactorisation tests: 18 passed
• benchmark output is generated by benchmarks/johnson_mercier.py

Stack

This is the generic Firedrake foundation. The existing simplex sum-factorisation PR will be retargeted to this branch. Paired FIAT generic PR: firedrakeproject/fiat#276.

AI assistance

OpenAI Codex was used to assist with implementation, refactoring, benchmarking, and drafting this PR. The human contributor remains responsible for understanding, validating, and maintaining the changes.