Extra configurability in `parallelizeAllLike`

[shmsong]

🚀 The feature, motivation and pitch

This is following up on #1828.

It'd be very helpful if we can add a few more configurability in parallelizeAllLike.

Currently the features I could think of are:

• An option to restrict what parallel type to propagate.
• A factory interface enabling actions like propagateAllParallelTypeExcept as well as propagateTheseParallelTypes etc.
• A rule to define regions where we want to propagate parallel type, on both the sub-graph of the tensor dag and the axis positions within a reference tensor. Similar to transform propagator.

Alternatives

No response

Additional context

No response