Partially fixes#15656 . When a crate graph is extended which is the case when new workspaces are added to the project
the rules for deduplication were too strict. One problem that arises from this is that in certain conditions
when we see the same crate having different `CrateOrigin`s the first form would be maintained. This approach however
results in some unwanted results such as making renaming forbidden as this has been recently only made available for
local crates. The given example in #15656 can still not be resolved with this PR as that involves taking inconsistencies
between dependencies into consideration. This will be addressed in a future PR.
internal: port anymap
## Description
- The anymap crate has been ported. During this process, unnecessary features for rust-analyzer have been removed.
- From the tests that were checking the existing licenses, the anymap license (`BlueOak-1.0.0 OR MIT OR Apache-2.0`) has been removed.
## Requests
- While porting the code this time, I have tried to respect the original author's intentions and have kept the comments/codes as much as possible. Please don't hesitate to tell me if you think the comments/codes also need to be appropriately modified.
- If there are any necessary changes regarding the licensing or anything else, please let me know so I can fix them.
## Issue
https://github.com/rust-lang/rust-analyzer/issues/15500
VSCode behaves strangely, allowing to navigate into label location, but
not allowing to apply hint's text edit, after hint is resolved.
See https://github.com/microsoft/vscode/issues/193124 for details.
For now, stub hint resolution for VSCode specifically.
Rationale: Merge commits will probably end up in
`rust-lang/rust-analyzer` when doing "rust=>ra" syncs anyway.
It could be changed to only check for merge commits in non-sync PRs,
but it's "probably not worth the hassle"
This adds an `in-rust-tree` feature that will be enabled when
rust-analyzer is built from `rust-lang/rust`. Due to the way
"git subtree" works, that test _will_ find merge commits and
fail, so we simply skip it.
The direct reason for this is to fix CI on windows, which seems to fail
for some reason after we fixed the watcher-selection logic which (I
think) changed the tests behavior to use notify rather than client.
But this patch seems to make sense in general -- file watching is
notoriously finicky, so controlling it explicitly leads to less fragile
tests.
