dialects: (linalg) add correct generic printing for fill, and matmul

[jorendumoulin]

I happened to need the fill operator and decided to do the matmul as well. This continues the work of #2959
I also made sure that the code works on more than just floating points values.
I moved some duplicate code to determine the type arguments of the hidden region into a common method of the NamedOp base class.
I also used the implicit builder to build the regions as I think this is much more readable.

Doing the matmul was a bit more tricky:

• for some reason this op prints the attributes in front of the ins and outs
• generic from includes the linalg.memoized_indexing_maps attribute, which are the indexing maps that would be generated when converting to a linalg

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 89.86%. Comparing base (27e0746) to head (32000bc).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2971      +/-   ##
==========================================
+ Coverage   89.84%   89.86%   +0.01%     
==========================================
  Files         408      408              
  Lines       51028    51053      +25     
  Branches     7912     7919       +7     
==========================================
+ Hits        45848    45878      +30     
+ Misses       3929     3925       -4     
+ Partials     1251     1250       -1     

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

[n-io]

Is this not just

for op in operands:
  op_type = op.type

?

[jorendumoulin]

Yes, this is probably somewhat cleaner

[n-io]

Very minor, but the op_type = op.type line was more to show equivalence rather than meant to be a concrete suggestion. Happy either way though!

[n-io]

Thanks, looks good!