Revert "Fix token usage / cost often being underreported"

[mrubens]

Reverts #6122

After using this for a bit, I agree with the previous comment that it's better UX to interrupt the request than to get the cost perfect.

---

Important

Reverts token usage reporting improvements to prioritize user experience by interrupting requests instead of perfecting cost reporting.

• Behavior:
  • Reverts changes from Fix token usage / cost often being underreported #6122, prioritizing user experience over precise token usage reporting.
  • Removes background task for usage data collection in Task.ts, opting to interrupt requests instead.
• Code Changes:
  • Deletes DEFAULT_USAGE_COLLECTION_TIMEOUT_MS from global-settings.ts.
  • Removes drainStreamInBackgroundToFindAllUsage function and related logic in Task.ts.
  • Adjusts recursivelyMakeClineRequests in Task.ts to handle request interruptions without background usage collection.

^{This description was created by for 015b876. You can customize this summary. It will automatically update as commits are pushed.}

[roomote]

Thank you for this revert! I've reviewed the changes and this is a clean revert that properly prioritizes user experience over perfect cost reporting. The decision to favor immediate interruption makes sense given user feedback. I have a few suggestions for consideration below.

[roomote]

Consider adding a comment here explaining the trade-off decision. Something like:

Suggested change

	// cost of the API cost not being retrieved.
	// PREV: We need to let the request finish for openrouter to
	// get generation details.
	// UPDATE: It's better UX to interrupt the request at the
	// cost of the API cost not being retrieved.
	// This prioritizes user experience (immediate interruption) over
	// perfect cost tracking based on user feedback.

This would help future developers understand why this approach was chosen.

[roomote]

The removal of the lastApiReqIndex < 0 check could potentially cause issues if this function is called with an invalid index. Consider keeping this defensive check even though the background processing is removed:

Suggested change

	const updateApiReqMsg = (cancelReason?: ClineApiReqCancelReason, streamingFailedMessage?: string) => {
	const updateApiReqMsg = (cancelReason?: ClineApiReqCancelReason, streamingFailedMessage?: string) => {
	if (lastApiReqIndex < 0 || !this.clineMessages[lastApiReqIndex]) {
	return
	}
	const existingData = JSON.parse(this.clineMessages[lastApiReqIndex].text || "{}")

[roomote]

The comment here specifically mentions OpenRouter, but this behavior affects all providers that return usage data at the end of streams. Consider making it more generic:

Suggested change

	// PREV: We need to let the request finish for openrouter to
	// PREV: We need to let the request finish for providers to
	// get complete usage/generation details.
	// UPDATE: It's better UX to interrupt the request at the
	// cost of the API cost not being retrieved.