feat: Add Expr.replace_strict support for lazy backends if default is provided

[FBruzzesi]

What type of PR is this? (check all applicable)

• 💾 Refactor
• ✨ Feature
• 🐛 Bug Fix
• 🔧 Optimization
• 📝 Documentation
• ✅ Test
• 🐳 Other

Related issues

• Related feat: Add default in {Expr,Series}.replace_strict #3276 (comment) and feat: Add default in {Expr,Series}.replace_strict #3276 (comment)

Checklist

• Code follows style guide (ruff)
• Tests added
• Documented the changes

If you have comments or can explain your changes, please do so below

Very first iteration, mostly a proof of concept to make sure there is a brute forced way for each backend.

I am not targeting main, rather #3276, so that I already have:

• correct signature elsewhere
• all the tests I need to check that the behavior is as expected

[FBruzzesi]

CI failures are interesting:

• pyspark succeeds, but pyspark connect doesn't. It might be worth reporting this upstream
• no coverage is because sqlframe does fail. I will make sure to open an issue with all the methods we are using that are currently missing

[FBruzzesi]

Alright, I reported the issue to SQLFrame and added the relevant link and comments in the code. Yesterday I learned that reporting spark issue is just not that simple, so I settled on xfailing for pyspark connect.

Regarding code sharing, I tried to abstract away a few parts but I didn't like the amount of inputs required to make it work, so I am marking this ready for review as is for now

[MarcoGorelli]

thanks @FBruzzesi !

I just simplified it slightly at the compliant level, as there were some unnecessary checks