Slightly less bad error messages for promote_shape

[kshyatt]

Having been enraged by these existing cryptic error messages, I tried to give the user a bit more information about what went wrong. Happy to take suggestions on further improvements.

[fredrikekre]

Fixes #33273 ?

[JeffBezanson]

👍
Something about DimensionMismatch("dimensions must match") certainly smells wrong. How about adding two fields to DimensionMismatch for the things that don't match?

[kshyatt]

Is that what we want to do? You can have three things which have a dimensionality issue, and this occurs in the LinearAlgebra stdlib. (Here, for example).

[JeffBezanson]

I guess it could be a tuple of N things. It can also be optional; we can still allow just passing a string. We do want to move away from interpolating in error messages though, since it is generally hard on the compiler (it makes many functions recursive, via a circuitous route through all of the I/O and printing code), makes it harder to compile things to e.g. GPUs, and it would often be better to separate the information from the presentation.

[kshyatt]

Yeah, that sounds reasonable. But IDK if I want to tackle rewriting DimensionMismatch entirely in this pr :).

[kshyatt]

BSD failure was in LibGit2.

[kshyatt]

Got the go-ahead from Jeff to merge this.