[Model] Add Qwen3 and allow switching between thinking and non-thinking mode

[CharlieFRuan]

This PR adds Qwen3 models from WebLLM 0.2.79.

We also add a config enable_thinking that can be toggled in the settings page to decide whether the Qwen3 model should reason or not. By default, it is turned off. Turning it on will cause it to reason and will also change its temperature and top_p, following https://huggingface.co/Qwen/Qwen3-0.6B#best-practices

In addition, when we want to summarize/create a topic, we should never reason, as it would take way too long.

In addition, for multi-round chat, we remove the <think>thinking tokens here</think>\n\n from the history assistant's message.

[CharlieFRuan]

@Neet-Nestor Hi Nestor! Could you review this PR when you get a chance? Currently, changing the setting page does not seem to be very robust. Specifically, I added an enableThinking checkbox, which when checked, we should see reasoning behavior in Qwen3 models. This only rarely works as expected for now. Thanks!

Edit: should be fixed now

[CharlieFRuan]

Never mind, the settings toggling should work now. There was a bug in isDifferentConfig(): config[field] !== config[field] should be this.llmConfig[field] !== config[field], as fixed in the latest commit.

[Neet-Nestor]

@CharlieFRuan I made several changes:

1. Put the thinking toggle button in model setting instead of global settings
2. Add another thinking toggle button in chat actions toolbar
3. Fix the issue of <think> </think> tags in chat titles
4. Update store version number in config.ts

I have tested and verified these changes by Qwen-0.6b. In the first screenshot the thinking toggle is on while in the second it's off.

[Neet-Nestor]

I made another change to remove the empty <think> </think> tags in non-thinking mode.

[CharlieFRuan]

@Neet-Nestor Thank you so much! These are great changes!