Fix issues with forwardDims()

To be merged after !90 (merged).

Fixes two issues with forwardDims():

• The previous iterative algorithm may be stuck in specific cases (see example below);
• It should always recompute all nodes output dims, because the graph may have changed.

Example of stuck case:

1. Initial graph is output dims forwarded;
2. Scaling operators are inserted before some Conv;
3. The graph is output dims re-forwarded. Here is what might happen:
   • Start the forward dims list with Producers;
   • Move to next child nodes: the Conv;
   • Conv cannot be forward dim because their new input, a Scaling node, is not forward dim;
   • Check if Conv childs are forward dim: if they are not directly the new Scaling node, they are!
   • We get stuck in a state where listNodes == nextList, although all nodes output dim are computable!

assigned to @olivierbichler

changed target branch from main to dev

changed title from Move computeOutputDims() function from GenericOp to OpTensor python binding so... to Fix issues with forwardDims()

added Important LanguageC++ PriorityHigh labels

changed the description

added 3 commits

• c8414a33 - 1 commit from branch dev
• 1a946b18 - Merge branch 'dev' into log
• caed37b6 - Merge branch 'log' into forwarddims

Compare with previous version

enabled an automatic merge when the pipeline for caed37b6 succeeds

approved this merge request

changed milestone to %aidge_core - v0.2.0

requested review from @pineapple

aborted the automatic merge because target branch was updated

added 2 commits

• 70729d7e - 1 commit from branch dev
• f998fb57 - Merge branch 'dev' into forwarddims

Compare with previous version

enabled an automatic merge when the pipeline for f998fb57 succeeds

mentioned in commit d00e9a7f

merged

mentioned in issue #48 (closed)