improve mixpanel usage statistics

[cebtenzzre]

Goals of this PR:

• Make it easier to add and modify events
• Better tracking of session duration
• Send events more frequently, especially within a single chat session
• Send more LocalDocs-related events

New events:

• ipify_complete: when the user's IP address has been determined (happens after startup now)
• response_complete: when the model has finished generating
• stop_generating_clicked: when the "stop generating" button is clicked
• regenerate_clicked: when the "regenerate" button is clicked
• send_message: when a message is sent via pressing Enter or clicking the send button
• session_end: when GPT4All is about to close
• doc_collection_add: when a LocalDocs collection is successfully added via settings
• doc_collection_remove: when a LocalDocs collection is removed manually or because the folder was deleted
• localdocs_indexing_start: when localdocs begins indexing any number of collections
• localdocs_indexing_complete: when localdocs finishes indexing any number of collections
• localdocs_folder_indexing: when localdocs begins indexing a folder that actually changed
• localdocs_folder_complete: when localdocs finishes indexing a folder that actually changed
• localdocs_startup: when localdocs is ready to start indexing at startup (for number of collections)
• opt_in: mirrors behavior of opt_out, to compute ratio of opt-ins to opt-outs

Removed events:

• chat_started (replaced by first property of response_complete)
• network_toggled (replaced by datalake_active property)

Changed event meanings:

• startup is now sent immediately after first run dialog is dismissed instead of after ipify is complete and a model is loaded, and will never be sent more than once per session
• opt_out will never be sent more than once per session
• model_load is now also sent for the first load event (before we were sending startup)

Changed event properties:

• all events: added session_id, removed requestedDevice (except for model_load), removed model and actualDevice (except for chat events), time is now in milliseconds instead of seconds
• chat events: added doc_collections_enabled, doc_collections_total, datalake_active, using_server
• startup: added $screen_dpi, datalake_active, cpu on Linux and Windows
• model_load: added $duration, error, model_arch (when unrecognized), cpu_fallback_reason

Other changes:

• Always display first start dialog if privacy options are unset (e.g. if the user closed GPT4All without selecting them)
• LocalDocs scanQueue is now always deferred
• Fix a potential crash in magic_match
• LocalDocs indexing is now started after the first start dialog is dismissed so usage stats are included