Log how much time loading a compiled artifact takes

[zou3519]

When people say they don't like the "torch.compile startup time", they mean two things:

1. the cold start time
2. the warm start time (when the vLLM disk cache has already been
   populated).

We had logging for (1), we didn't have (2). This PR adds (2)

Test Plan:

I ran VLLM_USE_V1=1 python benchmark_latency.py --model meta-llama/Meta-Llama-3-8B --batch-size 1 -O '{"level": 3, "compile_sizes": {1, 2}}'

And observed the following logs:

INFO 04-18 08:26:11 [backends.py:431] Dynamo bytecode transform time:
5.03 s
INFO 04-18 08:26:15 [backends.py:120] Directly load the compiled
graph(s) for shape None from the cache, took 4.190 s
INFO 04-18 08:26:18 [kv_cache_utils.py:634] GPU KV cache size: 532,032
tokens

Side note: it's probably not good that loading from the cache takes 4 seconds?

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀

[youkaichao]

the code change looks good to me. I didn't expect it takes so long though.