This refactors `collectAppPageSegments` to be more efficient especially
when dealing with multiple parallel routes, where the number of
combinations of segments can drastically increase how many segments are
returned per route. Previously this function only considered the
`children` parallel route. When it was updated to consider all possible
parallel routes in #73358, this had the unintended side effect of
increasing the amount of duplicate segments this function would have to
process.

We only need to return unique segments discovered from a particular
loader tree (representing a page), as opposed to the same segment as
many times as it might appear across all parallel routes. This PR uses a
map to track unique segments with a composite key used to identify the
discovered segments.

Additionally, this refactors the function to be iterative rather than
recursive. We keep track of a queue of segments to be processed (a tuple
of `[loaderTree, segments]`), and as we discover new parallel route
paths, we process the queue further.

Fixes #73793
Closes #73871 (h/t to @icyJoseph for identifying the cause of the memory
consumption)