fix: MemMapFs.Remove returns error for non-empty directories

[Yanhu007]

Fixes #232

Problem

MemMapFs.Remove() silently removes a non-empty directory and all its contents. This violates the os.Remove contract, which returns ENOTEMPTY for non-empty directories.

fs := afero.NewMemMapFs()
fs.MkdirAll("/dirname", 0755)
afero.WriteFile(fs, "/dirname/file", []byte("data"), 0644)
fs.Remove("/dirname") // succeeds, should fail with ENOTEMPTY

Fix

Before removing a directory, check if any entries have the directory path as a prefix. If children exist, return &os.PathError{Op: "remove", Path: name, Err: syscall.ENOTEMPTY}.

This matches the behavior of os.Remove on real filesystems.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[afurm]

The prefix check strings.HasPrefix(p, prefix) with prefix = name + FilePathSeparator correctly identifies children at one level. However, does the in-memory file system also support deep nesting like /dir/subdir/file.txt? If so, the check would also need to handle that the child path itself could have arbitrary depth. For a MemMapFs, the key question is whether a child at any depth under name should cause the remove to fail, or only direct children. The POSIX ENOTEMPTY typically applies to direct children only — but the current check also catches grandchildren since strings.HasPrefix("/dir/subdir", "/dir/") is true. Is that the intended behavior, or should the check be more precise?