SPMI: Avoid duplicate example diffs in diffs summary

[jakobbotsch]

We frequently see examples like this in the diffs report produced:

This PR deduplicates the contexts picked by method name, ensuring that we only get one of each name. I've also improved the logic to pick example regressions to show. After:

cc @dotnet/jit-contrib

[dotnet-policy-service]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

[tannergooding]

One other thing I've noticed in some diffs is that we'll only get diffs visible for the locals info dump and not actual code. It would be nice if that could also get resolved and to favor diffs that start after the first instruction group (in the actual code).

[jakobbotsch]

/azp run runtime

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[jakobbotsch]

One other thing I've noticed in some diffs is that we'll only get diffs visible for the locals info dump and not actual code. It would be nice if that could also get resolved and to favor diffs that start after the first instruction group (in the actual code).

I remember @EgorBo had the same wish when I initially added the examples. One simple thing we could do is just pick the last X lines of the diff instead of the first X lines.

The only thing is that the diffs in the local vars table can actually be pretty useful when looking at the examples, to make quick determinations. For example it quickly tells me about physical promotions or if strength reduction kicked in. But it makes sense to me to prefer the codegen diffs when the diff is too large to show fully.