Coalesce Oracle token estimate updates

[baron]

Summary

• Coalesce Oracle upcoming-input token estimate work instead of canceling/restarting every update.
• Snapshot the latest compose/conversation state after a short coalescing delay.
• Use the prompt view model's cached token total so the passive "Next msg" estimate no longer rebuilds full chat context/clipboard during typing or restore bursts.

Notes

• This intentionally trades some estimate precision for responsiveness when the chat preset's resolved context differs from the composer cached total, especially git-inclusion cases.
• No tests were added: Fable 5 review called out that testing this private UI-label coalescing path would require disproportionate refactoring.

Validation

• swiftformat --config .swiftformat Sources/RepoPrompt/Features/Chat/ViewModels/Oracle/OracleViewModel.swift
• make dev-swift-build PRODUCT=RepoPrompt
• swiftformat --config .swiftformat --lint Sources/RepoPrompt/Features/Chat/ViewModels/Oracle/OracleViewModel.swift
• swiftlint lint --strict --config .swiftlint.yml Sources/RepoPrompt/Features/Chat/ViewModels/Oracle/OracleViewModel.swift
• .agents/skills/rpce-contribution-check/scripts/preflight.sh commit
• .agents/skills/rpce-contribution-check/scripts/preflight.sh push

Review

• Claude Code Fable 5 (claudeCode:claude-fable-5) performed a final read-only minimization review and recommended GO after trimming redundant defensive cancellation machinery.

[provencher]

Actually oracle token counting isnt really used anywhere. We should just remove this altogether. it made sense in the IDE mode chat in classic to show the token count there. Here we dont really need this.