Switch from bazel_clang_tidy to aspect-build/rules_lint for clang-tidy validaton

[Databean]

This comes with some tradeoffs:

• It is possible to have different .clang-tidy files for subdirectories, as long as they are exported to the linting rule.
• clang_tidy invocation now happens at a build step, concurrent with other compilation operations.
• IIUC the actual validation happens in a test step, which reports success or failure based on the build step.
• The validation can be quite slow, even for moderate sized targets. One target in //cuttlefish/common/libs/utils took about 2 minutes. Based on CPU utilization, it seems like clang_tidy is being run serially for all source files in a build target, and only different build targets can be validated in parallel with each other.

[cjreynol]

on my machine:

• bazel clean then bazel build ... jumps from ~67 to ~190-200 seconds
• changing cuttlefish/host/commands/cvd/cli/commands/cache.cpp and triggering an incremental build jumps from ~3-4 seconds to 120 seconds

The jump in clean build times I am not so concerned about, but the incremental build jump is steep. Based on your last comment we could maybe split up some of our targets to get a speedup, but I would rather not make library structure decisions based on our tooling speeds.

A quick search shows we could use something like bazel query results piped to bazel build avoid test targets, but that also has downsides.

I did test using tags = ["avoid_me"], on the clang_tidy_test targets and running bazel build --build_tag_filters="-avoid_me" ... which brings the times back to the current numbers.

https://bazel.build/docs/user-manual#build-tag-filters)

I would prefer something like the tags option (or anything similar) to be able to keep the incremental builds speed down.

[Databean]

The incremental build time is still good for specific targets like bazel run //cuttlefish/host/commands/cvd -- ... but you're right that it is easy to walk into a slow case with something like bazel build //....

[cjreynol]

From our lunch conversations, I can just build the cvd target and skip the broad wildcard target.

[Databean]

I did test using tags = ["avoid_me"], on the clang_tidy_test targets and running bazel build --build_tag_filters="-avoid_me" ... which brings the times back to the current numbers.

Added the "clang_tidy" and "clang-tidy" tags to enable this strategy.

[Databean]

Thanks for reviewing!