Specifically, adds jake lssl (language service server library). The goal was to export the Session type which, in turn, has necessitated exporting most of the types in editorServices.ts by extension.

Additionally, I've made some changes to enable this library to function outside node - (node.d.ts is no longer compiled into the lssl target by any of the files included in it). Specifically, the two features of Node which Session used - Buffer.byteLength and process.hrtime, I've offloaded to a new Environment interface, and created a node implementation of it for tsserver - server\nodeimpl.ts. I was considering that they may be better provided by the System interface, and would like feedback on this - changing the System interface seemed like a farther reaching change than was desired.

On top of this, to make Session (or trivial subclasses) actually instantiate at runtime in non-node environments, I removed Session's meager dependency on ts.sys - which only gets set in node or in windows script host. For the most part, this simply meant calling the host which Session was constructed with, but in one instance this meant that the host had to be passed along to a static function - ScriptVersionCache.fromString. Additionally, the static CompilerService.defaultFormatCodeOptions had to be made aware that ts.sys may not be set at static initialization time if not running in either node or windows script host.

The remaining changes, b8f6ada and 9a3fead, are less critical to exporting the Session type as a library, but do provide quality improvements in the exported interface.

• b8f6ada allows consumers of the Session class to subclass and override onMessage and still call the onMessageParsed implementation - or just call onMessageParsed directly. This is useful in situations where the service is capable of receiving message objects as objects directly, rather than as json strings, thereby allowing consumers to remove (de)serialization overhead.
• 9a3fead adds support into the new onMessageParsed function to handle arbitrary messages, rather than throwing immediately if the massive switch statement falls all the way through. (IMO, it may be nicer to register all message handlers this way and remove the switch entirely.) This is incredibly useful in situations when the session communication channel is intended to the the primary communication channel between a client and a server, and allows a server to provide extended functionality (or, in the case of running in a browser worker thread, allows a client to send the server content with which to build a virtual filesystem and external edits to the files therein). This is the least required of the changes herein, so if too much issue is taken with it, it is not necessarily required if the prior commit is still accepted.