Hoist reduction outside a loop

[binarman]

This PR introduces an optimization that hoists reduction operation of dot accumulator outside a loop over K dimension:

%acc = for k tiles: %acc3d_input = reshape %acc %acc3d_out = dot3d(%x, %y, %acc3d_input) %acc = reduction batch %acc3d_out

transforms to:

%acc3d = for k tiles: %acc3d = dot3d(%x, %y, %acc3d) %acc = reduction batch %acc3d

This PR is a part of PR series. Final goal is to improve efficiency of small dot operations and bypass as much shared memory accesses as possible.

Rough list of PRs:

• [ ] Basic FMA dot fixes, dot 3d support and relaxing small dimensions for dot #4516
• [ ] Blocked->dotOp shared memory bypassing #4538
• [ ] Accelerate AMD Matmul + emit dot operations #4594
• [ ] Layout optimization, so operand B is loaded in proper mfma layout and do not need to go through LDS #4581
• [ ] Vectorization optimization of dot operands/results (in case llvm can not do this internally)
• [ ] Reduction operation hoisting out of the K loop (reduction operation is a byproduct of layout optimization step) (this PR) #4559