wasi: benchmarks fd_readdir for both real and fake file systems

[codefromthecrypt]

It is likely we'll have to special-case real files in wasi, as wasi-libc makes behavioral choices based on presence of non-zero inode data. This adds a defensive benchmark, so that we know what our performance base case is. This adds both real and fake file systems as both patterns are in use today.

$ go test -run='^$' -bench '^Benchmark_fdReaddir$'
goos: darwin
goarch: amd64
pkg: github.com/tetratelabs/wazero/imports/wasi_snapshot_preview1
cpu: Intel(R) Core(TM) i9-9880H CPU @ 2.30GHz
Benchmark_fdReaddir/embed.FS-16         	  457398	      2416 ns/op
Benchmark_fdReaddir/os.DirFS-16         	   45662	     26271 ns/op
PASS
ok  	github.com/tetratelabs/wazero/imports/wasi_snapshot_preview1	35.632s

See https://github.com/WebAssembly/wasi-sdk/releases/tag/wasi-sdk-17 and WebAssembly/wasi-filesystem#65

cc @jerbob92

[codefromthecrypt]

I added a benchmark of a continued read. This reads the same amount of dirents, but some from memory and rest from disk.

Benchmark_fdReaddir
Benchmark_fdReaddir/embed.FS
Benchmark_fdReaddir/embed.FS-16                    	  443401	      2460 ns/op	     384 B/op	       5 allocs/op
Benchmark_fdReaddir/embed.FS_-_continued
Benchmark_fdReaddir/embed.FS_-_continued-16        	  548738	      2173 ns/op	     288 B/op	       4 allocs/op
Benchmark_fdReaddir/os.DirFS
Benchmark_fdReaddir/os.DirFS-16                    	   44121	     27032 ns/op	    2648 B/op	      30 allocs/op
Benchmark_fdReaddir/os.DirFS_-_continued
Benchmark_fdReaddir/os.DirFS_-_continued-16        	  252295	      5047 ns/op	    2096 B/op	      21 allocs/op

[codefromthecrypt]

Thanks for the help again @jerbob92. I'll implement the fstat next, since we can see how much it costs now.