Implement local version of vdim

[tanderson92]

Experimental branch/PR to play around and track progress on ideas related to a local version of VDIM, which is given in schematic form below.

[codecov]

Codecov Report

Attention: Patch coverage is 2.37530% with 411 lines in your changes missing coverage. Please review.

Project coverage is 57.98%. Comparing base (33f7682) to head (2006455).

Files	Patch %	Lines
src/bdim.jl	0.00%	107 Missing ⚠️
src/vdim.jl	0.00%	103 Missing ⚠️
src/quadrature.jl	0.00%	69 Missing ⚠️
src/reference_interpolation.jl	0.00%	53 Missing ⚠️
src/mesh.jl	5.45%	52 Missing ⚠️
ext/IntiMakieExt.jl	5.26%	18 Missing ⚠️
src/api.jl	0.00%	6 Missing ⚠️
src/adaptive.jl	60.00%	2 Missing ⚠️
src/utils.jl	75.00%	1 Missing ⚠️

❗ There is a different number of reports uploaded between BASE (33f7682) and HEAD (2006455). Click for more details.

HEAD has 3 uploads less than BASE

Flag	BASE (33f7682)	HEAD (2006455)
	4	1

Additional details and impacted files

@@             Coverage Diff             @@
##             main      #82       +/-   ##
===========================================
- Coverage   79.55%   57.98%   -21.58%     
===========================================
  Files          25       25               
  Lines        2950     3342      +392     
===========================================
- Hits         2347     1938      -409     
- Misses        603     1404      +801     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[tanderson92]

Possible performance improvements left, taken from a quick look through profiling:

1. Vectorization of special functions calls using IntelVectorMath.jl or LoopVectorization.jl, in nystrom.jl
2. ( ✔️) Use of Bessels.jl for faster special function evaluation than SpecialFunctions.jl currently used
3. Avoid re-allocs for construction of Smat, Dmat, Vmat.
4. Avoid re-allocs for repeat in constructing sparse matrix
5. For local volume quadratures (currently called Yvol), extract these on the fly from parent quadrature rather than building anew. This is delicate if we wish to modify a QuadratureNode e.g. for re-centering and scaling the integration region.
6. Better neumann trace computation for $\gamma_1B$ -- a lot of time spent in getindex which seems wrong?
7. Remove duplication in computation in SingleLayerKernel and DoubleLayerKernel. Also, optimization of kernels themselves (e.g. r = x - y; d = norm(r) is not great). @fastmath too?

[tanderson92]

Feature (not performance) TODO before ready to merge:

• Fix stability via integration in $\tilde{x}$ coordinates; that is, handle homogeneous-preserving PDEs (those whose poly solutions are homog if RHS is homog, e.g. Laplace, Stokes) and non-homogeneous-preserving PDEs (e.g. Helmholtz) separately, the latter via kernel-split reduction to zero-freq limit PDE VDIM (i.e. for Helmholtz this is Laplace) in the low-freq limit.
• Vector-valued PDEs