Frame transform graph does not show the use of intermediate frames in transformations

As noted in #10867 during a discussion with @mhvk and @ayshih the transform graph is a little misleading since it implies, for example, that an ITRS->ITRS' self transform exists. In reality, there is a loopback through ICRS via intermediate frames...

One way of addressing this is to implement a hybrid of #8996 or #7207, so that frame transforms that don't exist are automatically routed through ICRS if possible, or raise an Exception if not.

If we don't like this for some reason, perhaps it would be possible to simply remove these from the graph? Or adding text to the loopback to indicate the frame that is ultimately responsible, like this?


pinging @eteq for comments

Start with the discussion in #10867 and compare the proposed approaches in #8996 and #7207. Trace how the transform graph represents ITRS→ITRS' and its loop through ICRS; done means agreeing on one behavior, implementing it consistently, and confirming that nonexistent transforms are routed or rejected as intended.

Written by the indexing model from the issue text.