first of all due to some technicalities we can no longer get_workbench - we shouldn't anyway (LoroDoc doesn't implement Clone so I can't just return a clone of the workbench itself)
it forced a long awaited fix: JS no longer duplicates the whole WASM memory
it should just hold the oplog (history) and evtree (LoroDoc) as well as a project and workbench index. that's it

so JS instead of calling Project::get_workbench now calls get_workbench_oplog and get_workbench_evtree where both require a project_index and workbench_index. both regular incremented numbers, nothing fancy
I also introduced set_workbench_evtree since we want a way for JS to update the evtree

so what changed?

• as explained above JS no longer acts on top of the lib.rs Project struct - it can be removed completely - but calls static global functions
• for that to be able to happen, JS has to keep track of the current project and workbench index (regular 0-based indexes)
• each workbench now holds an instance of the EvTree struct. A thin wrapper around LoroDoc
• the LoroDoc holds a list of step hashes. that's it. (this choice was made for us since Loro can't hold arbitrary types)
• step hashes are now unique (they take into account the author (hardcoded for now) and timestamp in addition to the message) so that we don't fall into collisions
• the hashes that LoroDoc holds is always a sub-group of the actual history. the lorodoc can fork or whatever but the history is always linear
• in order to compensate for the evtree vs actual workbench state getting out of sync (e.g. checking out before a sketch line gets created won't remove the line from the isotope sketch) a new method is introduced in the project::Project struct: rebuild_workbench. It accepts a workbench index and it:
  • nukes the workbench of all the internally held geometry BUT keeps the history and evtree intact
  • iterates through all the pointed evtree hashes (remember that loro abstracts away the complexity of the actual tree and gives us back the actual used history - from the currently checked out node to the root)
  • iterate through the workbench history to find the actual steps that the evtree hashes refer to
  • iterate through the found steps and call their handler (omitting the logic that appends them to the history and evtree)
• added methods to get the evtree and oplog
• added method to set the evtree

but there are quite a bit to get done in the UI:

• call the new global functions instead of wasmProject.*
• no longer hold a $workbench state but hold $oplog, $evtree, $project_index, $workbench_index and maybe $project (for the name?) instead
• some book-keeping is probably required for oplog and evtree (on send_message update them both)
• introduce any required loro-releated functionality for the evtree