[clangd] Always run preamble indexing on a separate thread

This has been the default in our production setup for weeks now,
showing great improvements to latency and no problems around stability or
correctness of the results.

Differential Revision: https://reviews.llvm.org/D155619