Reuse ProjectCachePlugin lookup state on cache miss

I've been experimenting with building an MSBuild accelerator using `ProjectCachePlugin`. I noticed that even a no-op plugin implementation that always returns `CacheMiss` and does no real work can still regress build time versus plain `dotnet build` by a noticeable amount. I'm seeing roughly **30-50%** in some runs, though I still need to gather more precise numbers.

One likely source is that MSBuild prepares project state for cache lookup before knowing whether the plugin will hit, then still performs a normal build on miss.

Current flow in `ProjectCacheService.PostCacheRequest`:

```text
EvaluateProjectIfNecessary(...)
-> LoadProjectIntoConfiguration(...)
-> configuration.Project.TranslateEntireState = true

new BuildRequestData(configuration.Project, targets)
GetCacheResultAsync(...)
```

If the plugin returns `CacheMiss`, the subsequent build should reuse this lookup-prepared project state and avoid any additional project load, translation, or transfer work beyond target execution.

Could MSBuild ensure that cache-lookup preparation fully substitutes for normal build preparation on `CacheMiss` / `CacheNotApplicable`?

Acceptance criteria: an all-miss no-op `ProjectCachePlugin` build should be close to the equivalent build without a cache plugin, aside from plugin callback/logging/scheduling overhead.

Related: #7112, #7121, #10615, #10616, #11490.

Start in ProjectCacheService.PostCacheRequest and trace EvaluateProjectIfNecessary, LoadProjectIntoConfiguration, and the BuildRequestData creation shown in the issue. Compare an all-miss ProjectCachePlugin build with an equivalent plain dotnet build; done means lookup preparation is reused on CacheMiss or CacheNotApplicable without extra project loading, translation, or transfer work beyond target execution.

Written by the indexing model from the issue text.