Fix supertrait wrapping when within max_width

[xonx4l]

This PR fixes an issue where supertraits were being forced to a new line even when they fit within the configured max_width as mentioned in #6809

Proposed Solution:
Use the original indented Shape for trait bounds instead of a manually reduced one preventing the "double-width reduction" math error in format_trait. This allows rewrite_assign_rhs to correctly calculate available horizontal space, keeping bounds on the same line when they fit within max_width.

The Changes -
Updated format_trait in src/items.rs to use the original indented offset and full available width for the bounds_shape.
Added rewrite_assign_rhs to handle the width reduction naturally based on the actual length of the trait header (lhs).
Replaced the manual newline logic with the dynamic wrapping provided by RhsAssignKind::Bounds.

[jieyouxu]

Question (for other rustfmt maintainers): I believe max_width is a stable option with a default value of 100.

• This impacts a stable option.
• This impacts formatting of the default value of a stable option.
• The stable code is already formatted.

I believe this fix needs to be gated behind a style-edition.

[xonx4l]

You are right . I think gating this behind StyleEdition::2024 will be good and won't regress formatting . Thanks!

[jieyouxu]

I think you will need to gate it behind StyleEdition::2027, StyleEdition::2024 is already the default inferred style edition if crate Rust edition is 2024.

[xonx4l]

Gated it behind StyleEdition::2027. Thanks!

[matthewhughes934]

To open a discussion: I also had a look into this issue and came up with another approach: c58a7c1 (but saw this was opened before creating a PR) that passes the test suite and so I don't think that change would require a version gate