feat: Support describe on decimal

[flisky]

fix #13732

[codecov]

Codecov Report

Attention: Patch coverage is 98.68421% with 1 lines in your changes are missing coverage. Please review.

Project coverage is 81.22%. Comparing base (5d449cc) to head (0cabf0b).
Report is 20 commits behind head on main.

Files	Patch %	Lines
...polars-core/src/chunked_array/ops/aggregate/mod.rs	98.11%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #15092      +/-   ##
==========================================
+ Coverage   81.06%   81.22%   +0.15%     
==========================================
  Files        1342     1346       +4     
  Lines      173935   175299    +1364     
  Branches     2459     2506      +47     
==========================================
+ Hits       141000   142386    +1386     
+ Misses      32468    32433      -35     
- Partials      467      480      +13     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[alexander-beedie]

I'll defer to others (@orlp?) but I don't think casting to f64 is a valid approach here 🤔 It's both an expensive conversion and, importantly, loses the point of using Decimal for calculations in the first place (the fixed-precision guarantees).

[orlp]

@alexander-beedie I think it's fine for describe.

[alexander-beedie]

@alexander-beedie I think it's fine for describe.

If it was constrained within the describe method then I'd agree, but the implementation appears to be available everywhere (describe uses select internally), eg:

df.select(
    sum=pl.sum("a"),
    min=pl.min("a"),
    max=pl.max("a"),
    mean=pl.mean("a"),
    median=pl.median("a"),
)

[flisky]

It turned out only mean need to be specialized for i128.
Please take a look, thanks!

[marcounternaehrer-cf]

Any chance to get this reviewed? :)

[flisky]

Rebased!

[ritchie46]

Thanks @flisky