Add reduction_size and broadcast_size attributes to XeTile.reduction and XeTile.broadcast

[Jianhui-Li]

Add broadcast_size to XeTile.reduction and XeTile.broadcast to support partial reduction.

We found two use case for partial reduction:

1. Support the staged work group reduction - first reduce within workgroup and then reduce across workgroup.
 %18 = math.exp %17#0 {map = #xetile.wg_map<sg_layout = [8, 4], sg_data = [32, 32]>} : vector<256x128xf32>
%19= vector.shape_cast %18 {map = #xetile.wg_map<sg_layout = [8, 4], sg_data = [32, 32]>} : vector<256x128xf32> to vector<8x32x128xf32>
%20 = vector.multi_reduction <add>, %19, %cst_0 {map = #xetile.wg_map<sg_layout = [8, 4], sg_data = [1, 32]>} [1] : vector<8x32x128xf32> to vector<8x128xf32>

        => 
 
 %18 = math.exp %17#0 {map = #xetile.wg_map<sg_layout = [8, 4], sg_data = [32, 32]>} : vector<256x128xf32>
 %20 = xetile.reduction <add>, %18, %cst_0 {map = #xetile.wg_map<sg_layout = [8, 4], sg_data = [1, 32]>} [1] {$reduction_size = [32]} : vector<256x128xf32> to vector<8x128xf32>

Support MXFP reduction - the reduction only reduce 32 elements to one element, so not reduce the whole dimension to one element.

This PR also fixes a few name inconsistency: tile_broadcast to broadcast, tile_reduce to reduction, tile_transpose to transpose, atomic_rmw_tile to atomic_rmw, tile_conv_layout to convert_layout.

Please review these guidelines to help with the review process:

• Have you provided a meaningful PR description?
• Have you added a test, a reproducer, or a reference to an issue with a reproducer?
• Have you tested your changes locally for CPU and GPU devices?
• Have you made sure that new changes do not introduce compiler warnings?
• If this PR is a work in progress, are you filing the PR as a draft?
• Have you organized your commits logically and ensured each can be built by itself?

[nbpatel]

LGTM. Can we also change xetile.tile_conv_layout to xetile.conv_layout

[Garra1980]

do we need to mention vector<2x64xfloat> twice in this operation?

[Jianhui-Li]

fixed. thanks