transformations: Add memref-stream-tile-outer-loops

[superlopuh]

We sometimes have too high a rank when lowering to snitch_stream, this avoids that at the memref_stream level, before streamification.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 89.85%. Comparing base (05b2fdf) to head (d1ce389).
Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2869      +/-   ##
==========================================
- Coverage   89.87%   89.85%   -0.02%     
==========================================
  Files         396      397       +1     
  Lines       49827    49950     +123     
  Branches     7683     7712      +29     
==========================================
+ Hits        44780    44883     +103     
- Misses       3837     3848      +11     
- Partials     1210     1219       +9     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[zero9178]

Regarding tests: Really just the most basic kernel would be fine to use here that happens to have more than the given rank of dimensions being used. Having at least one test case where the pattern has to be applied twice (i.e. cut two dimensions in total, one at a time) would be interesting.

Additionally if you address one of the comments: A test where the input memrefs are already strided with an offset (possibly static)