Remove dynamic memory allocations inside lowest level assembly functions

[garth-wells]

This will allow the lowest lever assembly functions to be executed efficiently for small batches of cells.

Also makes a number of simplifications following careful review of code.

[michalhabera]

It would be also neat to check no heap allocation in certain scopes. Clang (>=19) has [[clang::nonallocating]] attribute, https://clang.llvm.org/docs/FunctionEffectAnalysis.html#the-nonblocking-and-nonallocating-attributes Requires extra -Wfunction-effects pass.

[schnellerhase]

That flag could be passed to the clang-tidy linting run in https://github.com/FEniCS/dolfinx/pull/3722. Clang tidy hooks into a LLVM based compile anyways.

[garth-wells]

Will update and merge in smaller pieces.