257 verify loop fusion

[codecov[bot]]

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Project coverage is 99.86%. Comparing base (70f9793) to head (1a275a4).

Additional details and impacted files

```diff @@ Coverage Diff @@ ## master #2707 +/- ## ======================================= Coverage 99.86% 99.86% ======================================= Files 354 354 Lines 49082 49137 +55 ======================================= + Hits 49018 49073 +55 Misses 64 64 ```

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

[hiker]

I've found an additional problem with the loop fuse: the loops can be specified in the wrong order, and this will be accepted by validate. But during fuse the loop bodies will be ending up in the wrong order!

[hiker]

Ready for a first review. I have moved the dependency analysis required in the validation function into the DependencyTool , anticipating the need to use the existing code for DA. Unfortunately (and as discussed), I don't have time atm to implement the full DA for fusion, I only verify that all accesses in the fused dimension are consistent (i.e. no stencil read access for variables that are also written) - this can result in some cases to reject fusion when it is still valid. But we have a 'force' option so we can overwrite this (and much better to reject a bit too keen compared with the current state where we incorrectly fuse).

Note that in order to support renaming of loop variables I had to add support to SymPy to indicate that some variables have the same value.

I also moved most of the failing loop fusion tests into the dependency tools test (to reach 100% coverage of the latter).

I've triggered the CI as well.

[hiker]

Ready for next review. I've triggered the CI, since I'd guess there is a good chance for this change affecting some transformations.