Round floats but not decimals in SqlLogicTests

[findepi]

Which issue does this PR close?

Closes #.

Rationale for this change

• stop rounding decimal values in SLT. It's the very nature of decimal
  arithmetics that it should in general be exact.
• round float values taking into account the bit width of the float. The
  Float64 carries more information than Float32 or Float16.

What changes are included in this PR?

• SLT formatting

Are these changes tested?

yes

Are there any user-facing changes?

no

[findepi]

cc @gliga

[findepi]

Bummer. We get types different from PostgreSQL, so it's hard to get results rendered the same if rendering is type dependent.

[Omega359]

I think this will impact me significantly with the sqlite test file integration I'm working on.

[findepi]

in a positive or negative manner?

[Omega359]

in a positive or negative manner?

Unsure @findepi. Below is an example of a variety of discrepancies between DataFusion and Postgresql results as an example of what I'm trying to deal with (df on the left, pg on the right)

[findepi]

float rounding could help with some of these eg

However, the reason why i didn't finish this PR is that even for existing modest number of tests this is a problem.
This is because we get types different. For example 2.0 is apparently decimal in PostgreSQL, while it's a float in DF. Maybe addressing #12817 and this PR would make landing sqlite slt tests easier?

[Omega359]

float rounding could help with some of these eg

However, the reason why i didn't finish this PR is that even for existing modest number of tests this is a problem. This is because we get types different. For example 2.0 is apparently decimal in PostgreSQL, while it's a float in DF. Maybe addressing #12817 and this PR would make landing sqlite slt tests easier?

the slt tests are in place now. We could run them with your changes and see what breaks and whether it's a correct breakage or not.