REF: minor performance improvements in Graph

[martinfleis]

Follow up on #691 and conversion to neighbors and weights dicts as discussed there.

I am also changing the sparse format to CSR which is what we need for lag and generally anything else and what weights is using. That also allows cleaner implementation of _components.

Plus using matrix_power implementation for scipy sparse array (courtesy of @ljwolf) which allows us to skip conversion to sparse matrix.

Note that the change of the sparse representation is technically a hard-breaking change but given we're in experimental phase we can afford it.

[codecov]

Codecov Report

Attention: Patch coverage is 96.55172% with 1 lines in your changes are missing coverage. Please review.

Project coverage is 85.0%. Comparing base (bcabdbc) to head (438cf5f).
Report is 3 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##            main    #697     +/-   ##
=======================================
- Coverage   85.0%   85.0%   -0.0%     
=======================================
  Files        141     141             
  Lines      15203   15223     +20     
=======================================
+ Hits       12924   12941     +17     
- Misses      2279    2282      +3     

Files	Coverage Δ
libpysal/graph/_spatial_lag.py	100.0% <100.0%> (ø)
libpysal/graph/tests/test_base.py	100.0% <100.0%> (ø)
libpysal/graph/base.py	97.7% <96.0%> (-0.2%)	⬇️