resolve TempDir paths before passing them to the flake bin

[valkum]

On macOS /tmp is a symlink to /private/tmp.
TempDir creates a dir and returns the path as /tmp/.... nix is unhappy with symlink dirs when calling nix flake metadata --json path:...

error (ignored): error: end of string reached
error:
       … while fetching the input 'path:/tmp/colmena-assets-XHHAXJ'

       error: path '/tmp' is a symlink
thread 'main' panicked at src/cli.rs:286:38:
Failed to get flake or hive: ChildFailure { exit_code: 1, backtrace:    0: backtrace::capture::Backtrace::new
   1: colmena::nix::flake::FlakeMetadata::resolve::{{closure}}
   2: colmena::nix::flake::Flake::from_uri::{{closure}}
   3: colmena::cli::run::{{closure}}
   4: tokio::runtime::park::CachedParkThread::block_on
   5: colmena::main
   6: std::sys_common::backtrace::__rust_begin_short_backtrace
   7: std::rt::lang_start::{{closure}}
   8: std::panicking::try
   9: std::rt::lang_start_internal
  10: _main
 }
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

To fix this we canonicalize the path first, resolving any symlinks on the way.

[valkum]

Regarding the CI test failures: These are also present on main!

[Mic92]

I found a few usages of this: #236

[Mic92]

Not sure if those are relevant, but my thinking was that it might resolve other future regressions.

[valkum]

I did this near the point where the data (the temp path) leaves our control as this isn't a bug in the Rust side of things but at the boundary to a different system.
This only affects paths that we pass to nix (so ideally we would canonicalize the paths in FlakeMetadata::resolve). We could do that by passing something like a FlakePath to that function instead of the str.

enum FlakePath<'s> {
Path(&'s Path),
Git(&'s str)
}

<Display impl that canonicalizes before creating the `path:{}` string>

This would allow us to keep the simple temp dir paths in places where it could get debug printed without causing any irritation why the tmp dir looks weird. The symlinks for tempdirs on macos are an implementation detail and should not be exposed in level-1 debug output (such as a dbg!()) print, IMHO.

[zhaofengli]

Tested that it fixes evaluation on macOS. Will add a test to keep it working later.