Follow conjugation convention in tree_dot

[Randl]

It may break something downstream unfortunately, so updating dependencies should be done carefully

[patrick-kidger]

Modulo failing tests -- not sure what's going on with those (?) -- this LGTM.
Once the tests are fixed we can check the Optimistix and Diffrax tests on this PR before merging it, and see if anything looks awry.

[Randl]

Should we bump the version since this is a potentially breaking change?

[patrick-kidger]

Yes, I think we should.

I can see that tests now pass -- are you happy with this PR / should I merge it?

[Randl]

I think I am.
Though one possibility to consider is to not do conjugation in tree_dot at all but rater leave it to caller, similarly to jnp.dot. This may clutter downstream code but will be consistent with the Jax/numpy semantics.

[patrick-kidger]

Merged, then!
And that's an interesting idea. In particular given the difficulties we faced in patrick-kidger/optimistix#61, I can see that that might simplify things.

The decision is yours -- I'd be happy to make that change if you think that is what is best.