Implement dev live optimizer

[spalger]

Fixes: #4569
Fixes: #4554
Fixes: #4556
Fixes: #4555
Fixes: #4549

This pr offers the the following changes:

• in dev mode, use version of the compiler that:
  • watches files for changes
  • only rebuilds when needed by delaying rebuilds until files are requested by the browser
• only use babel to compile .babel.js files, dramatically shortening webpack build time
• switch to prebuilt moment.js without locale support
• out of dev mode, bundles are correctly reused
• restored the loading screen from before Apps/App Switcher #4335

There are also a number of new config options available, each with the goal of giving users more control over how things are built.

• optimize.bundleDir : the directory that bundles and related assets should be written.
  • default: '{kibanaRoot}/optimize/bundles'
• optimize.bundleFilter : a glob that selects which bundles to build. Other bundles are ignored and might as well not exist
  • default: * unless optimize.tests is true
• optimize.sourceMaps : Defines what type of source maps should be created.
  • default: true in dev, false otherwise
  • options:
    • true: produce high quality maps with line and column precision
    • false: disable source maps completely
    • "fast": make quick and dirty maps (currently alias for cheap-module-eval-source-map)
    • Anything that webpack's devTool option accepts
• optimize.unsafeCache : Enables a caching mode that may cause the optimizer to not notice added or removed files. Using this option greatly improves build performance, but is opt-in so that users know to look out for invalid cache hits.
  • default: '/[\/\\](node_modules|bower_components)[\/\\]/'
  • options:
    • true
    • false
    • a string wrapped in / used to create a RegExp
• optimize.lazyPrebuild : When starting up the dev server, run the optimizer once before initializing the server.
  • default: true
• optimize.profile : Profile the build and upon successful completion write a stats.json file into the optimize.bundleDir that can be used with the webpack analyze tool to analyze build performance and contents
• optimize.tests : Create the test bundle and nothing else, combine with optimize.bundleFilter to build the tests along with other bundles
  • default: false

The primary motivator for this change was the speed of the feedback loop. My testing offers the following stats:

Optimization times:

./bin/kibana

no cache

~18s

./bin/kibana --optimize.bundleFilter=appSwitcher

no cache

~8s

./bin/kibana --dev

initial	~28s
refresh	~7s

./bin/kibana --dev --optimize.unsafeCache=true

initial	~25s
refresh	~3s

./bin/kibana --dev --optimize.unsafeCache=true --optimize.sourceMaps=false

initial	~18s
refresh	~1s

./bin/kibana --dev --optimize.unsafeCache=true --optimize.sourceMaps=false --optimize.bundleFilter=appSwitcher

initial	~8s
refresh	~0.25s

Memory usage:

dev
coordinator	~105 MB
server	~150 MB
optimizer	~550 MB
production with single optimization run
server	~600 MB
production without optimization run
server	~150 MB
production post #4597 without optimization [1]
server	~80 MB

CPU Usage still climbs to 100% during optimization, but afterward the processing again idle at (or near) 0

[1] All threads are currently compiling the server's source code with babel at require-time which results in a slight performance penalty and memory overhead, but with #4597 we should be able to optimize this away.

[w33ble]

Could do with less newlines here...