Prototype "lazy" haste resolver

[afoxman]

Create a prototype "lazy" haste resolver as described in #992.

Run experiments on small, medium and large monorepos and a unirepo, gathering performance data. Measure speed gain/loss (clock time), memory peak, I/O (bytes read/written). Compare normal watchman runs with "lazy" runs.

Report all findings. If viable, develop an engineering plan to implement in rnx-kit.

[afoxman]

Related to jestjs/jest#12231

[afoxman]

Metro recently released their jest-haste-map impl as a proper package: facebook/metro#812

[tido64]

As a temporary workaround, we've implemented a way for reducing the scope of the Haste resolver. It's currently implemented as a monkey patch since there is no good way to "configure" this behaviour in Metro. Pending internal testing, we will look at ways to upstream this work: #2044

[kelset]

we want to start Metro without incurring in the cost of Metro's initial crawling