Restore {Code,TracebackEntry}.path to py.path and add alternative

[bluetech]

In 92ba96b we have changed the path
attribute to return a pathlib.Path instead of py.path.local without
a deprecation hoping it would be alright. But these types are somewhat
public, reachable through ExceptionInfo.traceback, and broke code in
practice. So restore them in the legacypath plugin and add Path
alternatives under a different name - source_path.

Fix #9423.

[bluetech]

The alternative to this as discussed in the issue is to go through with the breaking change.

For this PR:

• Honor our promise to not break stuff without deprecation
• Will restore compat for the promise project and whatever other unknown projects that will be broken by this

Against this PR:

• source_path is uglier and less discoverable than path
• We estimate not many users are using this at all
• Some code might work fine with the silent switch from py.path -> pathlib and will end up better
• The path fix is temporary as we aim to deprecate and remove py.path stuff eventually

Overall it will be better to go with the breaking change in this case, i.e. close this PR and document the change as breaking rather than trivial as 92ba96b did. WDYT?

[bluetech]

This part also fixes the sybil failure - it passes a string absolute path to cut, which used to work because py.path.local('/foo') == '/foo' but Path('/foo') != '/foo'. So the code now compares strings.

Even if we go with the breaking change option we should get this included in pytest 7.0 (this part is non breaking)

[nicoddemus]

Overall it will be better to go with the breaking change in this case, i.e. close this PR and document the change as breaking rather than trivial as 92ba96b did.

As I commented in the issue, I think we should just go for it, but I'm not against if others prefer to play safe here and go with the PR instead. 👍

[bluetech]

Replaced by #9447.