[GPU] Expand matmul decomp cases

[kealan-barbieri]

Description

Expands cases where inputs are decompressed to include integer activations as well as weights. Integer weights are still required but in the presence of fpmath setting activations will also be upconverted.

This means that cases intending to use integer accumulation must not supply eg attr-fpmath=f16:true as all such cases must be upconverted per https://jira.devtools.intel.com/browse/MFDNN-13380.

Align behavior with doc for application of fpmath to integer inputs, if integer fpmath is true then upconversion must occur, this changes the strategy selection. Relabel strategies that perform upconversion to make them explicit. Update strategy selection to distinguish ordering between different "compound" types eg "[OH]" "[OB]".

Fixes # (github issue)

Checklist

General

• [x] Do all unit and benchdnn tests (make test and make test_benchdnn_*) pass locally for each commit?
• [x] Have you formatted the code using clang-format?

[kealan-barbieri]

make test disable device_cpu enable device_gpu disable benchdnn_all enable benchdnn_matmul

[kealan-barbieri]

make test disable device_cpu enable device_gpu disable benchdnn_all enable benchdnn_matmul

[kealan-barbieri]

make test disable device_cpu enable device_gpu disable benchdnn_all enable benchdnn_matmul enable benchdnn_ip

[kealan-barbieri]

make test disable device_cpu enable device_gpu disable benchdnn_all enable benchdnn_matmul enable benchdnn_ip

[kealan-barbieri]

make test perf-gpu set primitive=matmul

[kealan-barbieri]

make test perf-gpu set primitive=matmul

[kealan-barbieri]

make test perf-gpu set primitive=matmul

[kealan-barbieri]

rendered unnecessary.