[mlir][vector] Linearize vector.create_mask (flatten to 1D)

[newling]

Alternative to #138214

Rather than using an arith.select, this PR computes a new rank-1 mask argument directly with scalars before splatting. This PR also supports more mask shapes.

FYI @nbpatel I started trying to suggest this approach in #138214 but thought it'd be clearer to just implement it. Please let me know what you think

[nbpatel]

why do we need an alternative for PRs already up for review? My PR already got a LGTM and some feedback. I guess if there is some feedback you have we can always discuss or you can review and leave feedback and I can incorporate it in my PR, also, you can create a iterative PR later to incorporate the changes you require. I don't feel duplicating work is the correct way forward

[newling]

Hi @nbpatel I don't intend to pursue this PR (it is a draft, so I don't expect review of it) but as I mentioned in the summary I thought it'd be quicker to describe my thoughts on #138214 by trying it code first. I intended to add comments in #138214 summarizing this, my apologies for not being clearer