GitLab

This change limits the number of hosts that are requested
requested by the prediction model fetcher from the remote
optimization guide service and is finch controllable. This will
improve the request latency and load on the remote optimization guide
service. It will also limit the total number of host model features
maintain on the client. Additionally, only hosts that are not already
in the host model features cache will be included in the request.

This change also implements a least-recently-used in-memory cache for
the host model features that the prediction manager maintains.
The maximum size is controllable via finch and the default is set
to hold 95% of all users hosts without needing to evict any host model
features.

Bug: 1036399
Change-Id: Ice9ab7c3c3505fc970ab72e1c78f5d628cd5157b
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1981632
Commit-Queue: Michael Crouse <mcrouse@chromium.org>
Reviewed-by: Tarun Bansal <tbansal@chromium.org>
Reviewed-by: Sophie Chang <sophiechang@chromium.org>
Cr-Commit-Position: refs/heads/master@{#727838}