Refactor benchmark taxonomy

Nobody has claimed this yet.

This ticket proposes a new taxonomy of benchmarks for GL JS which aspires to be comprehensive, based on measurements directly relevant to our end users, and conceptually neat.

# Performance Metrics

- time-to-`dataend` against {Chrome, Firefox, Safari, IE11, Edge}, several styles, several camera positions, and {initial style load, `#setFilter` operation, small `setData` operation, large `setData` operation}
- {95th precentile frame time, 95th percentile frame duration, mean frame time, mean frame duration, mean framerate} against {Chrome, Firefox, Safari, IE11, Edge}, several styles, several camera positions, and {panning, rotating, pitching, zooming within an integer zoom level, zooming across integer zoom levels}
- memory usage against {Chrome, Firefox, Safari, IE11, Edge}, several styles, several camera positions, and {panning, rotating, pitching, zooming within an integer zoom level, zooming across integer zoom levels}
- number of GL API calls against {Chrome, Firefox, Safari, IE11, Edge}, several styles, several camera positions, and {panning, rotating, pitching, zooming within an integer zoom level, zooming across integer zoom levels}
- power usage against {Chrome, Firefox, Safari, IE11, Edge}, several styles, several camera positions, and {panning, rotating, pitching, zooming within an integer zoom level, zooming across integer zoom levels} (we could create a lightweight native app with a webview to measure this)

# Other Metrics

* test coverage
* statements
* branches
* functions
* lines
* flow type coverage
* compressed & gziped bundle size
* CI build time
* codebase size (we can use nyc for this too)
* statements
* branches
* functions
* lines
* number of flattened npm dependencies
* score measuring platform parity (GL JS vs GL Native vs Leaflet)

The issue names no files, tests, or benchmark entry points. Begin by locating the existing GL JS benchmark infrastructure and comparing it with the listed performance and project metrics; completion would require an agreed, implementable taxonomy.

Written by the indexing model from the issue text.