[Fix] Allow concurrent inference for multi model in WebWorker

[CharlieFRuan]

This is a follow-up to #542. Update examples/multi-model to use web worker, and to also show case generating responses from two models concurrently from the same engine. This is already supported for MLCEngine prior to this PR, but WebWorkerMLCEngine needed a patch. Specifically:

• Prior to this PR, WebWorkerMLCEngineHandler maintains a single asyncGenreator, assuming there is only one model.
• Now, to support concurrent streaming request, we replace this.asyncGenerator with this.loadedModelIdToAsyncGenerator, which maps from a model id to its dedicated asyncGenerator
• As a result, messages related to streaming need to specify the selectedModelId, hence the updates for the message sending and handling of chatCompletionStreamInit, completionStreamInit, completionStreamNextChunk
  • Because upon handling these messages, the handler needs to know for which model to initiate an async generator, or on which async generator to call .next() on
  • This also means completion() and chatCompletion() of WebWorkerMLCEngine will call getModelIdToUse(), which prior to this PR delays till the underlying MLCEngine
• As of now, this.loadedModelIdToAsyncGenerator may not be cleaned properly when one asyncGenerator finishes. We only call clear at unload(), which may not be called upon reload(). However, service_worker may skip reload(). Will leave it as is for now.

Tested with WebLLMChat, also tests WebLLMChat terminating service worker manually.

[CharlieFRuan]

Demo of examples/multi-models with parallelGeneration():

web-llm-multi-models.mov