Fix spacing issues with deepseek models:

[manyoso]

1. If the think tag is empty don't show it at all
2. Always trim the responses for whitespace so if the model generates
   blah
   [new line]
   final response

The finale response is trimmed of the newline.

[cebtenzzre]

If we trim leading whitespace here, then we no longer need to do so in onRegularResponse:

-auto respStr = QString::fromUtf8(m_result->response);
-return onBufferResponse(removeLeadingWhitespace(respStr), 0);
+return onBufferResponse(QString::fromUtf8(m_result->response), 0);

I think the reason we don't currently remove trailing whitespace while generating is that it causes a "jump" where if the model outputs e.g. 10 blank lines and one new word, it will only show any change once it generates the word.

Is there any reason related to tool calling that we would need to start doing that here?

[manyoso]

Maybe we should just trim leading whitespace here?

[manyoso]

see new version. taking the temporary is gross though

[cebtenzzre]

I don't mind; it's just like glib's g_strchug (or C functions like strtok).

[cebtenzzre]

While we're here, shouldn't this be more like:

    auto respStr = finalBuffers.size() > 1 ? finalBuffers.last() : QString::fromUtf8(result.response);
    if (!respStr.isEmpty() && std::as_const(respStr).back().isSpace()) {
        m_chatModel->setResponseValue(respStr.trimmed());
        emit responseChanged();
    }

The purpose of this code is to trim the trailing whitespace from the end of the response if needed, since the response handler won't do it while generating. I don't see why any check on result.response is necessary if we aren't using its value.

[manyoso]

Yeah, could be simplified.