feat!: Split replace functionality into two separate methods

[stinodego]

Closes #14302

Changes

• Add replace_strict method to Expr and Series. This method is similar to what replace was previously, with the exception that it raises an error if any non-null values were not replaced. Specify a default to set all values that were not replaced to that default.
• replace now always keeps the existing data type - this is a breaking change (see example below). Thedefault and return_dtype parameters are deprecated.

Example

Before:

>>> s = pl.Series([1, 2, 3])
>>> s.replace(1, "a")
shape: (3,)
Series: '' [str]
[
        "a"
        "2"
        "3"
]

After:

>>> s.replace(1, "a")
Traceback (most recent call last):
...
polars.exceptions.InvalidOperationError: conversion from `str` to `i64` failed in column 'literal' for 1 out of 1 values: ["a"]
>>> s.replace_strict(1, "a", default=s)
shape: (3,)
Series: '' [str]
[
        "a"
        "2"
        "3"
]

[codspeed-hq]

CodSpeed Performance Report

Merging #16921 will not alter performance

_{Comparing replace-strict (2dc0d1f) with main (69b8440)}

Summary

✅ 37 untouched benchmarks

[codecov]

Codecov Report

Attention: Patch coverage is 95.95960% with 8 lines in your changes missing coverage. Please review.

Project coverage is 80.88%. Comparing base (8a6bf4b) to head (2dc0d1f).

Files	Patch %	Lines
crates/polars-ops/src/series/ops/replace.rs	96.80%	4 Missing ⚠️
py-polars/src/lazyframe/visitor/expr_nodes.rs	0.00%	2 Missing ⚠️
crates/polars-plan/src/dsl/function_expr/mod.rs	80.00%	1 Missing ⚠️
crates/polars-plan/src/dsl/mod.rs	95.65%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #16921      +/-   ##
==========================================
+ Coverage   80.86%   80.88%   +0.02%     
==========================================
  Files        1456     1456              
  Lines      191141   191297     +156     
  Branches     2728     2734       +6     
==========================================
+ Hits       154562   154727     +165     
+ Misses      36073    36064       -9     
  Partials      506      506              

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

[stinodego]

@ritchie46 This PR is now ready for review. Design is a bit different from what we discussed in the office:

1. Replacing nulls by a different value is supported. Reasons I went for this:

• We were already supporting it.
• It would be weird if replace(1, None) works but replace(None, 1) doesn't.
• It may be more efficient to replace everything in one go, rather than calling replace followed by fill_null.

2. Nulls from the original Series do not propagate to the output. This really doesn't make sense, especially when the user species a full column as the default.

[ritchie46]

Not for this PR, but I think we can set this to map later. I don't believe this is group-aware. If the input is group-aware, that would already bubble up that info.

[ritchie46]

For a future PR we must optimize this. Adding a validity is almost free, but here it takes a very expensive branchy path.

[stinodego]

Yeah I wasn't happy about this either - but I couldn't find a way to do this without a lot of code... I figure we're probably missing some API building blocks here.

[ritchie46]

There is still a few things left to optimize, but let's ensure we have the correct behavior for 1.0 first. 👍