Chore!: Refactor sqlmesh test output to use rich

[VaggelisD]

Before this PR:

• Default:

• Verbose

After this PR

• Default

• Verbose (columns are split pairwise)

[izeigerman]

Shouldn't a single -v be enough for the pairwise mode? Actually, I wonder whether we should just automatically switch to the pairwise mode if the number of columns exceeds certain threshold. Ideally, this should depend on the current width of the screen. Why even output something we know is probably unreadable?

[VaggelisD]

**Shouldn't a single -v be enough for the pairwise mode? **

Yeah, that is already the case too, both -v and -vv will cause the pairwise mode.

Actually, I wonder whether we should just automatically switch to the pairwise mode if the number of columns exceeds certain threshold. Ideally, this should depend on the current width of the screen. Why even output something we know is probably unreadable?

That is a good idea, but do we think this might cause too much output and render the default version useless? Maybe a different idea would be to break the initial table apart every N columns, e.g:

> sqlmesh test 

Data mismatch:
col_1: Expected | col_1: Actual | ... | col_n: Expected | col_n: Actual|

Could become:

> sqlmesh test 

Data mismatch (chunk 1):
col_1: Expected | col_1: Actual | ... | col_5: Expected | col_5: Actual|

Data mismatch (chunk 2)
col_6: Expected | col_6: Actual | ... | col_10: Expected | col_10: Actual|

....

What do you think?

[izeigerman]

That is a good idea, but do we think this might cause too much output and render the default version useless? Maybe a different idea would be to break the initial table apart every N columns, e.g:

I think I like that

[georgesittas]

LGTM, though, I'd let Iaroslav and Sung take another look at this. Looks nice 👍