Update matmul() to compute with posit arithmetic for posit32 dtype.

[xman]

Currently, posit32 data is converted to double, process, and converted back to posit32.