Copy entrypoints that have a shebang that differs in python vs python3

[zanieb]

In #14919 it was reported that uv's behavior differed after the first invocation. I noticed we weren't copying entrypoints after the first invocation. It turns out the shebangs were written with .../python but on a subsequent invocation the sys.executable was .../python3 so we didn't detect these as matching.

This is a pretty naive fix, but it seems much easier than ensuring the entry point path exactly matches the subsequent sys.executable we find.

I guess we should fix this in reverse too? but I think we might always prefer python3 when loading interpreters from environments.

See #14790 for more background.

[woodruffw]

Looks right to me!

To confirm my understanding, here's the old behavior:

# OK, rewritten correctly
# previous_executable: /some/random/python3
#!/some/random/python3

# BUG, not rewritten correctly
# previous_executable: /some/random/python3
#!/some/random/python

Whereas the new behavior would rewrite both of these to /some/new/python3, correct?

[zanieb]

Yep! Thanks!