bug: dry run mode behaviour not align with actual run

[VioletsOleander]

Consider the following example:

Previous recnys.yaml:

{
    "foo/",
}

Current recnys.yaml:

{
    "foo/bar/",
}

Then run re, the operations will be:

1. Unlink foo
2. Link foo/bar

Those are ideally correct.

But run re --dry-run, the operations will be:

1. Unlink foo
2. Error: path foo/bar is occupied, please remove it manually and try again

This is because the file existence check for dry-run mode and normal mode both
relies on the true file system tree on the disk, therefore, since in dry-run mode
operation 1 does not actually remove the symlink foo, operation 2 will still consider
foo/bar is an existent directory so that report an error.

The solution should be further isolate the logic of file system detection for dry-run
mode and normal mode, i.e. let dry-run mode maintain a fake file system in the memory.

Mark this as TODO.

[VioletsOleander]

A rather 'elegant' way is to introduce an additional filesystem abstraction (similar to pyfakefs), allowing unifying the execution layer.

The drawback is that I have to modify so many Path.xxx invocation.

Directly using pyfakefs as runtime dependency is not ideal, because it just mocks everything, and the concequence is not predictable. Also, when pyfakefs become a test dep and runtime dep simultaneously, testing dry-run mode will be werid.

[VioletsOleander]

Given that --dry-run flag is such a usually used flag, maybe implementing more logic for it is worthy.