save import hash info to state

[dberenbaum]

This saves the hash info from imports to the output state db so that we don't rehash imports. Without this PR, imported files were always hashed when the output was saved. Existing imported files were even hashed again when doing operations like dvc import -f.

This should improve the speed of all imports, and it also reverts the changes in #10388 that caused slowdowns in some scenarios. That previous PR wrongly assumed the slowdown occurred during download, when it actually slowed during output.save(). That PR happened to work in some scenarios because the workaround updated the state db.

[skshetry]

I did some minor adjustments to the PR (removed fs.walk() and instead, we get the results directly from fs.download(), also makes it cleaner as we don't have to handle path manipulation).

I also ran the dvc-bench benchmark for test_import in https://github.com/iterative/dvc-bench/actions/runs/10466916928/job/28984684233#step:9:44.

Name	(time in s)
test_import-import[import-skip-hash-2]	278.6997 (1.0)
test_import-import[main]	392.8382 (1.41)

This is now back to 2.58.2 level.

[dberenbaum]

Great changes, thanks! I think if we had a benchmark using large files, you would see an improvement over 2.58.2.