feat: add experimental support to run tasks via Nx

[vsavkin]

Nx is the third option in addition to p-map and p-queue

Description

Currently running lerna run will result in Lerna either delegating to p-map or p-queue to schedule tasks/processes. Lerna has direct dependencies on both p-map and p-queue.

The industry has advanced since the time of p-map and p-queue. Tools like Nx and Lage are much more powerful and efficient task orchestrators. They support computation caching and, in case of Nx, config-free task distribution. A lot of teams are moving away from Lerna for this reason.

This PR adds Nx (which is a MIT-licenced extensible build system) as a third option (in addition to p-map and p-queue). If you have the nx package installed, and you have nx.json in your workspace, Lerna can now delegate its lerna run commands to Nx. This is optional: you have to opt into it. And it's done in a backwards-compatible way: every lerna run command will work but will be significantly faster and can be distributed across multiple machines without any config.

How Has This Been Tested?

We have tested on several small Lerna repos. For instance, this repo contains a benchmark measuring cache restoration of different monorepo-based tools: https://github.com/vsavkin/large-monorepo

With this change enabled, Lerna is on par with Nx, and is significantly faster than other tools.

Once the beta with this change is out, we are going to add it on several large enterprise Lerna repos to measure the perf and make sure the change handles all edge cases well.

[juristr]

Just couple of observations