gh-155596: Fix pprint expand mode ignoring width for nested values

[cobed95]

Issue

• Issue: pprint(..., expand=True)'s "best effort" for adhering to width=... seems not very "best" #155596

Summary

In the new "expand" mode, 'key': prefix needs to be accounted for.
Here we add a new argument prefix_len to _format, and

1. In "expand" mode, subtract it from max_width so that width is respected.
2. In "aligned" mode, fold it into indent so that existing behaviour is preserved.

Tests

One new test was added and two existing tests' expectations have been updated.
Existing expectations were violating the width=40 setting, so I would like to regard
these updates not as regressions, but rather as corrections.

• Issue: pprint(..., expand=True)'s "best effort" for adhering to width=... seems not very "best" #155596

[python-cla-bot]

All commit authors signed the Contributor License Agreement.

[cobed95]

@hugovk @corona10 Hi sprint mentors, I would like to request reviews from you :)
Thanks a lot.

[hugovk]

_format is "private", but we should be careful that PrettyPrinter can still be subclassed by others who won't have prefix_len, so we'd want to avoid:

TypeError: PrettyPrinter._format() got an unexpected keyword argument 'prefix_len'

We've run into issues like this before. Is there a way to fix this without adding prefix_len here?

[cobed95]

@hugovk You're right, this would break subclasses.

I've reworked it to hand the prefix width over out of band instead. _format() keeps its existing signature and reads self._pending_prefix_len, which a new _format_child() sets immediately before the call and clears in a finally. The four call sites go through that helper, so _child_indent() is used as before and the diff gets smaller too.

A subclass that overrides _format() then just doesn't get the width correction.

One thing I wasn't sure about: I made _pending_prefix_len a class attribute so a subclass that skips super().__init__() still has a default. Happy to move it into __init__ if you'd prefer it explicit.