feat: add token-based condensing threshold support

[roomote]

Summary

This PR adds support for configuring the Condensing Trigger Threshold in tokens (in addition to the existing percentage-based configuration).

Changes

• Modified the sliding window logic to support token-based thresholds
• Values above 100 are now treated as token counts rather than percentages
• Token thresholds take precedence when both percentage and token thresholds could apply
• Added comprehensive tests for the new token-based threshold functionality

How it works

• Percentage thresholds: Values between 5-100 continue to work as percentages of the context window
• Token thresholds: Values above 100 are treated as absolute token counts
• Special value -1: Continues to mean "inherit from global setting"

Example Usage

Users can now configure profile-specific thresholds like:

• 50 = Condense when context reaches 50% of the window
• 8000 = Condense when context reaches 8000 tokens (regardless of window size)

This provides more precise control for profiles with known models and token limitations.

Testing

• Added 3 new test cases for token-based thresholds
• All existing tests pass without modification
• Tests verify that token thresholds trigger correctly and take precedence over percentage thresholds

Fixes #7440

---

Important

Adds token-based condensing thresholds to truncateConversationIfNeeded() in index.ts, with tests in sliding-window.spec.ts.

• Behavior:
  • Adds token-based condensing thresholds in truncateConversationIfNeeded() in index.ts.
  • Values >100 are treated as token counts, taking precedence over percentage thresholds.
  • Special value -1 continues to mean "inherit from global setting".
• Testing:
  • Adds tests for token-based thresholds in sliding-window.spec.ts.
  • Tests ensure token thresholds trigger correctly and take precedence over percentage thresholds.
• Misc:
  • Updates logic to handle both percentage and token thresholds, with token thresholds taking precedence.

^{This description was created by for be1511a. You can customize this summary. It will automatically update as commits are pushed.}

[roomote]

Reviewing my own code is like debugging in a mirror - everything looks backwards but the bugs are still mine.

[roomote]

Would it be helpful to add a comment here explaining the threshold logic? Something like:

Suggested change

	// Determine the effective threshold to use
	// Determine the effective threshold to use
	// Values interpretation:
	// -1: inherit from global setting
	// 5-100: percentage of context window
	// >100: absolute token count
	let effectiveThreshold = autoCondenseContextPercent

This would make it immediately clear to future maintainers how the different threshold values are interpreted.

[roomote]

Should we consider adding an upper bound check for token thresholds? For example, values larger than typical context windows (e.g., > 1,000,000) might indicate a configuration error.

[roomote]

The warning message could be more informative about the new token-based thresholds:

Suggested change

	console.warn(
	console.warn(
	`Invalid profile threshold ${profileThreshold} for profile "${currentProfileId}". Valid values are: -1 (inherit), 5-100 (percentage), or >100 (token count). Using global default of ${autoCondenseContextPercent}%`,
	)

[roomote]

Would it be worth adding a comment explaining this approximation? The 4 characters per token ratio is a testing simplification that future maintainers might find helpful to understand.

[roomote]

Consider extracting the magic number 100 as a constant like PERCENTAGE_TOKEN_BOUNDARY. This would make the distinction between percentage and token thresholds clearer throughout the code.

[daniel-lxs]

Closing this PR since it doesn't properly implement a token-based context threshold. The UI element to change the value for the percentage is a slider that only goes up to 100%. To properly implement this feature, the user should be able to switch between percentage and token modes explicitly, rather than using the ambiguous heuristic of treating values >100 as token counts. This approach would require UI changes to add a toggle or dropdown for selecting the threshold type.