Dialects: (csl) Simplify DSD types in IR

[n-io]

Simplify DSD types in IR by not carrying around element type information. Since this is only needed for the printing of @increment_dsd_offset(), we can provide a helper method instead to retrieve this information.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 89.67%. Comparing base (6ef4760) to head (56e6e00).
Report is 4 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2683      +/-   ##
==========================================
+ Coverage   89.64%   89.67%   +0.02%     
==========================================
  Files         364      364              
  Lines       46779    46871      +92     
  Branches     7090     7104      +14     
==========================================
+ Hits        41937    42031      +94     
+ Misses       3740     3739       -1     
+ Partials     1102     1101       -1     

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

[AntonLydike]

Not sure I'm 100% on board here, doesn't this allow us to type-confuse an integer dsd for a float one and create "bad" code?

[dk949]

I think it's important to keep in mind that the type here is only used to denote size, using i32 should be the same as using f32 (no matter the underlying type).

If a DSD was a void *, it'd be equivalent to doing dsd = ((char *)dsd) + sizeof(T)