Windows Daemon should respect DOCKER_TMPDIR

[ryansimmen]

- What I did
Fixes #35076 by instructing the windows daemon to utilize the tmp directory under data-root and respect DOCKER_TMPDIR if it exists.

- How I did it
Leveraged Windows TMP and TEMP environment variables.

- How to verify it

1. Install Docker for Windows
2. Pull a large image
3. Observe disk activity via Resource Monitor

- Description for the changelog
Bug fix: Windows Daemon will now utilize the tmp directory under data-root and respect DOCKER_TMPDIR if it exists instead of leveraging the Temp directory under C:\Windows.

[darstahl]

@jhowardmsft @johnstep PTAL

hcsshim uses Golang's built in tmpdir utilities, which create a tmp dir at the env TMP location during layer extraction. This doesn't currently respect Moby's tmpdir settings.

This would also fix any other dependencies using Golang's built in tmpdir utilities.

Note that Golang will not attempt to create TMP if it does not exist (layer extraction will fail in this case). We might want to add a check here to either bail out if TMP doesn't exist, create it, or not set TMP if DOCKER_TMPDIR does not exist.

[lowenna]

LGTM. But agree with @darrenstahlmsft that it might be good to stat the directory/create it if it doesn't exist, or error out if it doesn't exist. I don't think it matters which though.

[ryansimmen]

@jhowardmsft the code will now create the directory if it does not exist and error out if it cannot create it.

[yongtang]

LGTM

[PatrickLang]

We need to get this into a new EE release. Azure VMs are deploying with a 30GB C:, but 100GB D: . Many customers are putting DOCKER_ROOT on d: because there is more space available, its faster & cheaper. d: is local SSD storage so its perfect for non-persistent container storage

[thaJeztah]

@PatrickLang I can add an internal tracking issue for that, but catch up on slack internally to discuss prioritisation