Add --collect-only-tree flag for tree-style collection output

[sclaughl]

Closes #14150

Adds a new --collect-only-tree (--co-tree) flag that displays collected tests as a tree with box-drawing characters,
providing a more readable alternative to the classic <Type name> format.

pytest
└── testing
    └── test_example.py
        └── TestFoo
            └── test_bar

Features:

• Box-drawing characters (ASCII fallback for non-UTF-8)
• Color-coded output (directories/files bold, classes cyan, functions green)
• Respects -q (summary only) and -v (shows docstrings)

The existing --collect-only behavior is unchanged.

[nicoddemus]

Thanks @sclaughl for the PR. It seems very complete and well written.

However, I'm personally hesitant to add yet another flag (and a bunch of code to maintain) regarding this feature.

In my opinion we have 2 options:

1. Decline the PR; while it looks good, I don't think it is worth in terms of extra flags and maintenance to the core.
2. Replace the output from --collect-only with this new output; not sure it is worth the hassle, as this breaks backward compatibility (I'm sure there are scripts out there which parse the output of --collect-only).

What other maintainers think?

[RonnyPfannschmidt]

Its tricky the tree display is not meant for parsing so replacing is acceptable

The key details of reorder vs compact need some bikeshedding

[sclaughl]

I'd be happy to replace --collect-only entirely and remove the angle-bracket code (and tests). That would address the maintenance concern better than adding a new flag.

To be explicit about the design: the tree shows structural organization (by directory/module/class) and would not reflect reordering from pytest_collection_modifyitems. The hook would remain for use by features like --lf / --ff, but --collect-only would answer "how are tests organized?" rather than "what order will they run?"

For anyone who needs execution order, --co -q already shows nodeids in run sequence (sans angle brackets).

This seems like a good trade-off -- a proper tree can't represent arbitrary reordering without breaking its structure. Just wanted to flag it clearly since it's a behavior change.

Happy to rework the PR if you'd like to go in that direction.