[robhogan]

With Metro's defaults - lazy SHA-1s, no hasteImplModulePath and enableGlobalPackages: false - no file needs a worker at startup. Even so, every crawled file goes through FileProcessor to find that out: a check for node_modules in its path, a call to the Haste plugin's filter, which always says no, and an array for the plugins to run.

This makes HastePlugin.getWorker() return null when the plugin has nothing to compute, so it gets no worker and no per-file data slot (as MockPlugin already does). processBatch then returns straight away when there's no SHA-1 to compute and no plugin runs on batches.

Any plugin after Haste moves down a slot in each file's metadata, so this bumps CACHE_BREAKER.

Benchmark

This takes 13-14% off applying the crawl on a cold start on macOS and Linux (Linux: 84ms -> 73ms), or 4-5% off a cold start on Linux and Windows (Linux: 315ms -> 298ms). macOS's cold starts are too noisy to show it.

On Linux and Windows, a cache write takes one of two times (on Linux, about 14ms or 20ms), and this PR's writes take the shorter one more often (Linux: 12 of 20 runs against 5, Windows: 20 against 8). That's presumably because there's less garbage to collect, but it makes the cache write medians below jumpy. On macOS, writes are 0.1ms slower.

	main	#2219	This PR	vs main	vs #2219
macOS: Applying the crawl in a cold start	18.7ms	17.6ms	15.1ms	-18.9% (-20.7% to -17.3%)	-14.2% (-16.2% to -12.5%)
Linux: Applying the crawl in a cold start	85ms	84ms	73ms	-14.1% (-16.4% to -10.2%)	-13.2% (-15.5% to -9.7%)
Windows: Applying the crawl in a cold start	47ms	42ms	40ms	-15.5% (-18.1% to -11.8%)	-6.2% (-13.1% to +0.1%)
macOS: Cache write in a cold start	7.8ms	4.4ms	4.5ms	-41.9% (-42.7% to -40.3%)	+3.5% (+2.6% to +4.6%)
Linux: Cache write in a cold start	33ms	20ms	14.0ms	-57.8% (-59.0% to -42.0%)	-30.5% (-32.3% to -5.1%)
Windows: Cache write in a cold start	20ms	13.8ms	9.4ms	-54.2% (-55.9% to -52.6%)	-32.1% (-34.8% to +2.1%)
macOS: Cold start	129ms	95ms	91ms	-29.8% (-38.6% to -24.2%)	-4.7% (-16.7% to +4.8%)
Linux: Cold start	646ms	315ms	298ms	-53.9% (-54.6% to -53.5%)	-5.4% (-7.2% to -4.6%)
Windows: Cold start	362ms	172ms	164ms	-54.7% (-55.7% to -53.7%)	-4.5% (-5.7% to -2.2%)

Changelog

 - **[Performance]**: metro-file-map: Skip the per-file worker pass when no file needs a worker

Test plan

New unit tests. Benchmark as in #2104, with 22 rounds on macOS.