wasi: move Readdir logics from fdReaddirFn to Readdir methods

[evacchi]

As suggested in this comment, the logic that populates the fields in Readdir is moved to methods of the Readdir struct. We are not exporting an interface yet.

• The fields are no longer exported.
• There are new fields that are necessary to maintain the position of the cursor, and func fields to avoid embedding a FileEntry.
• We export 2 constructors: one takes a FileEntry, and it is defined in terms of the other; the other takes 2 funcs dirInit and dirReader that are invoked when the initializing, resetting, advancing the cursor and filling the internal buffer.

Notably, for simplicity, entries are read in batches; dirents is always at most large direntBufSize = 16 (currently), which is usually the (fixed) size of each batch, unless it is the initial fetch, or it is the last fetch:

• the initial fetch is dirReader(n), where n := direntBufSize - len(initial), and len(initial) is the size of the slice returned by dirInit()
• all fetches may return len(batch) <= direntBufSize; if len(dirent) < direntBufSize, then we have reached the end of the listing for this directory.

The cursor/iterator-like API has Peek(), Advance() methods. It also provides Skip(n), Reset(), Cookie() and Rewind(cookie).

• Advance() advances the cursor field and the countRead field. The countRead is the absolute index of the current item; cursor is the index in the current window (dirent)
• Rewind(cookie) allows seeking backwards to the position of the given cookie. The cookie corresponds to the internal countRead. When the cookie is 0, Rewind(0) is equivalent to Reset(). It is not possible to seek forward (beyond the cursor). It is not possible to seek backwards to a value outside the current window.
• Skip(n) is mostly used for testing, the other methods are more useful and used in fdReaddirFn too, and it is a short-hand for invoking Advance() n times.

Tests have been migrated to unit tests of Readdir when it made sense, and they were kept in the wasi namespace when it was simpler.

[evacchi]

thread 'main' panicked at 'assertion failed: `(left == right)`
  left: `0`,
 right: `1`: expected one entry', src/bin/fd_readdir.rs:167:5

eh fair enough 🤷 😛

[codefromthecrypt]

Definitely getting cleaner! let's make sure we retain sliding window so that a huge directory doesn't OOM. Ack it is more work, but worth it.

[evacchi]

Sure, will do the sliding window (in fact a few commits earlier it was there), but I would keep the window fixed (direntBufSize) instead of using the user request value, that should make it simpler. Unless we have a reason for that too

[evacchi]

restored the sliding window, cleaned up, squashed, rebased

[codefromthecrypt]

so much cleaner than what I had before. Great job!