Linear algebra type stability enhancements and matrix solve support

[Nader-Rahhal]

related Legate PR: JuliaLegate/Legate.jl#84
Currently uses 1 gpu

requires legate_branch: develop

[krasow]

@Nader-Rahhal I got the tests passing in CI; however, we need to expand the tests to encapsulate all supported types for solve and the lin alg test suite. Basically adding a for T in SUPPORTED_NUMERIC_TYPES (or being more restrictive for things that fail in solve) over the test case.

[Nader-Rahhal]

@Nader-Rahhal I got the tests passing in CI; however, we need to expand the tests to encapsulate all supported types for solve and the lin alg test suite. Basically adding a for T in SUPPORTED_NUMERIC_TYPES (or being more restrictive for things that fail in solve) over the test case.

Sounds good. I can get this implemented tomorrow.

[Nader-Rahhal]

@krasow I added the type coverage. For all types that arent FP and Complex, cupynumeric casts to f64 so we just copy that

[krasow]

@Nader-Rahhal You should use one of the predefined consts (defined in src/cuNumeric.jl) instead of listing out everything. Additionally, you can enable verbose=true on the testset so it prints out by T pass/fail.

[krasow]

I removed the as_type conversions to get solve passing for integer types. We can not support ints on solve. Or we can do an implicit type conversion to floats.