Back off token recount selection retries

[baron]

Summary

• Add a one-shot, capped retry backoff for token recounts when selection changes during a heavy token calculation
• Reset retry backoff on fresh selection projection changes and after a consistent recount
• Add DEBUG-only coverage for retry delay gating, cap, and reset behavior

Validation

• .agents/skills/rpce-contribution-check/scripts/preflight.sh commit ✅
• .agents/skills/rpce-contribution-check/scripts/preflight.sh push ✅
• swiftformat --config .swiftformat --lint Sources/RepoPrompt/Features/Prompt/ViewModels/TokenCountingViewModel.swift Tests/RepoPromptTests/MCP/MCPSelectionReplyFreshnessTests.swift ✅
• swiftlint lint --strict --config .swiftlint.yml Sources/RepoPrompt/Features/Prompt/ViewModels/TokenCountingViewModel.swift Tests/RepoPromptTests/MCP/MCPSelectionReplyFreshnessTests.swift ✅

Notes:

• make dev-lint was attempted but failed in the full-repo SwiftFormat lint phase with an unrelated environmentEntry rule timed out error in Sources/RepoPrompt/Infrastructure/WorkspaceContext/Models/WorkspaceCodemapLiveOverlayModels.swift.
• Focused XCTest compile was previously attempted but did not reach test execution before being canceled due a very long fresh worktree compile.

[sentry]

Bug: The selection consistency retry backoff can be canceled by concurrent non-selection dirty events (e.g., prompt text changes), defeating the purpose of the backoff delay.
_{Severity: MEDIUM}

Suggested Fix

The shouldUseSelectionConsistencyRetryBackoffForNextSelectionRecount flag should not be reset until the backoff-delayed task actually begins execution. Alternatively, when scheduleTokenCountUpdateIfNeeded() is called, it should check if the pending tokenUpdateDebounceTask is a backoff task and avoid canceling it for non-critical dirty events.

Prompt for AI Agent

Review the code at the location below. A potential bug has been identified by an AI
agent. Verify if this is a real issue. If it is, propose a fix; if not, explain why it's
not valid.

Location:
Sources/RepoPrompt/Features/Prompt/ViewModels/TokenCountingViewModel.swift#L286-L293

Potential issue: A race condition exists where the selection consistency retry backoff
is defeated. When a retry is scheduled, the
`shouldUseSelectionConsistencyRetryBackoffForNextSelectionRecount` flag is consumed and
a long debounce task is started. If a concurrent non-selection event (e.g., prompt text
change) triggers `markDirty()` before the backoff task executes, the pending backoff
task is canceled and replaced with a standard, shorter debounce task. This undermines
the feature's goal of reducing rapid recounts during periods of instability, leading to
degraded performance.

_{Did we get this right? 👍 / 👎 to inform future reviews.}