Fix infinite recursion in ignore

[wmccrthy]

ignore was recently ported over from a different repo, that used an adhoc path implementation. This repo's path.dirname returned a string or nil, but lute's path.dirname always returns a string.

On various occasions in ignore ( one example here), we gate continued processing behind checks that path.dirname(arbitrary) exists. These are all invalid checks now and in some of these cases, we recur leading to death by stack overflow.

This PR fixes aforementioned checks, using path.basename (returns string | nil if at root) to determine if the current path is the root.

I found this bug when encountering a stack overflow error while trying to use lute transform:

_readIgnoreRecursive
... (+19979 frames)
@cli/transform/lib/ignore.luau:227 function _readIgnoreRecursive
@cli/transform/lib/ignore.luau:227 function _readIgnoreRecursive
@cli/transform/lib/ignore.luau:227 function _readIgnoreRecursive
@cli/transform/lib/ignore.luau:227 function _readIgnoreRecursive
@cli/transform/lib/ignore.luau:227 function _readIgnoreRecursive
@cli/transform/lib/ignore.luau:214 function isIgnoredDirectory
@cli/transform/lib/files.luau:16 function traverseDirectoryRecursive
@cli/transform/lib/files.luau:52 function getSourceFiles
@cli/transform/init.luau:139 function main
@cli/transform/init.luau:153

I rebuilt lute with these changes and transform worked as expected.

[Vighnesh-V]

small change, but otherwise looks good to me.

[skberkeley]

we have path.path objects, so I believe the following would be more performant:
if #path.parts > 0 then path.dirname(filepath) else nil

[wmccrthy]

If filePath is always a table here, won't path.basename mirror this exact logic (non-nil if #path.parts > 0 and nil otherwise)? Happy to change but curious what the performance bottleneck would be

[skberkeley]

yeah, the logic will be the same, but we avoid the table check and passing the string basename around this way