try parallelizing independent barrier procedures

[chriscoey]

in particular:

• cone barrier procedures (one per primitive cone)
• for WSOS cones specifically, the n+1 indepedent cholesky factorizations
• ~~linear system solves (two independent sub-solves)~~ LAPACK does it for us now

these are the most expensive operations

[chriscoey]

waiting on PARTR to work nicely with BLAS and LAPACK multithreading (https://github.com/JuliaLang/julia/issues/32786)