perf: use fs.walk instead of fs.walkStream for async provider

[mrmlnc]

What is the purpose of this pull request?

Improving the performance of the fg('**') method.

Something like ~20% for medium directories and trees. Practically nothing for large directories and trees (1 million and more).

// *
===> Benchmark pattern "*" with 200 launches (regression, async)
===> Max stdev: 7 | Retries: 5 | Options: {}

Name                   Time, ms  Time stdev, %  Memory, MB  Memory stdev, %  Entries  Errors  Retries
---------------------  --------  -------------  ----------  ---------------  -------  ------  -------
fast-glob-current.js   6.970     0.402          4.388       0.014            5        0       1
fast-glob-previous.js  8.571     0.430          4.531       0.015            5        0       1

// **
===> Benchmark pattern "**" with 200 launches (regression, async)
===> Max stdev: 7 | Retries: 5 | Options: {}

Name                   Time, ms  Time stdev, %  Memory, MB  Memory stdev, %  Entries  Errors  Retries
---------------------  --------  -------------  ----------  ---------------  -------  ------  -------
fast-glob-current.js   89.665    3.884          9.649       0.380            11882    0       1
fast-glob-previous.js  112.966   5.370          7.760       0.112            11882    0       1

What changes did you make? (Give an overview)

Instead of using the ReaderStream reader, now we use the ReaderAsync reader. This avoids the overhead of stream processing.

But we still use ReaderStream for static method.