Don't make assumptions about Metadata based on temp_dir

[mrkline]

Previously, exa would attempt to get the metadata for
std::env::temp_dir, and if fetching a certain timestamp
(e.g., creation time) failed, it would assume that the OS couldn't
produce that stamp for any file. Requests to display or sort by that
timestamp would misfire the entire program.

This caused exa to make some sweeping claims that aren't true; e.g., that
Linux can't produce file creation times. The real story is that tmpfs
(the filesystem usually used for /tmp) can't, but other Linux
filesystems can and do. (The Rust standard library docs even mention
that Metadata.created() is backed by statx() since Linux 4.11.)

Since exa could be asked to list files from several filesystems,
let's handle metadata on a per-file basis. This patch makes File's
*_time() methods return an Option
(changed_time() always returns None on a non-Unix system),
and prints a hyphen for files without that particular timestamp.
For example:

    $ exa -lhUd --sort=created /tmp/example Cargo.toml Cargo.lock src target

    Permissions Size User    Date Created Name
    .rw-r--r--     0 mrkline -            /tmp/example
    .rw-r--r--   25k mrkline 12 Jan 22:35 Cargo.lock
    .rw-r--r--  1.2k mrkline 12 Jan 22:35 Cargo.toml
    drwxr-xr-x     - mrkline 12 Jan 22:35 src
    drwxr-xr-x     - mrkline 12 Jan 22:46 target

[mrkline]

Ah, apparently using statx() in the standard library is relatively new: rust-lang/rust#65094

[ogham]

Thanks for your contribution, and for pointing me towards statx — it's been a while since I looked at the stat manual page and that function was new to me.

In this case, I merged another PR that took care of the temp_dir call, so I'm afraid I can't merge this one.