For use cases with a large number of threads or a large number of isolates (or
both), ThreadDataTable can be a major performance bottleneck due to O(n)
lookup time of the linked list. Switching to a hash map reduces this to O(1).

Example 1: Sandstorm.io, a Node.js app that utilizes "fibers", was observed
spending the majority of CPU time iterating over the ThreadDataTable.
See: https://sandstorm.io/news/2016-09-30-fiber-bomb-debugging-story

Example 2: Cloudflare's Workers engine, a high-multi-tenancy web server
framework built on V8 (but not Node), creates large numbers of threads and
isolates per-process. It saw a 34x improvement in throughput when we applied
this patch.

Cloudflare has been using a patch in production since the Workers launch which
replaces the linked list with a hash map -- but still global.

This commit builds on that but goes further and creates a separate hash map
and mutex for each isolate, with the table being a member of the Isolate
class. This avoids any globals and should reduce lock contention.

Bug: v8:5338
Change-Id: If0d11509afb2e043b888c376e36d3463db931b47
Reviewed-on: https://chromium-review.googlesource.com/1014407Reviewed-by: Yang Guo <yangguo@chromium.org>
Commit-Queue: Yang Guo <yangguo@chromium.org>
Cr-Commit-Position: refs/heads/master@{#52753}