✨ add --recent-window option to limit burn rate calculation window

[guhyun9454]

Problem

When a user is idle for an extended period and then starts actively using Claude Code, the burn rate projection is diluted by the idle data points. Because calculate_burn_rate() applies linear regression over the entire window, a 2-hour idle period followed by 30 minutes of heavy usage results in a projection that significantly underestimates the current burn rate.

Solution

Add a --recent-window <duration> option (e.g. 30m, 1h) to all subcommands. When set, the burn rate regression only uses snapshots from the last N minutes, so the projection reflects the current usage pattern rather than a historical average.

ccburn session --recent-window 30m
ccburn weekly --recent-window 1h

Default is None — fully backward compatible with existing behaviour.

Implementation

• calculator.py: calculate_burn_rate() accepts recent_window_minutes and tightens the snapshot cutoff to max(window_start, now - recent_window). The min_span check is also scaled to the recent window size (not the full window) so short windows are not unfairly rejected.
• display/layout.py: BurnupLayout.update() threads the parameter through to calculate_burn_metrics().
• cli.py: RecentWindowOption added; reuses the existing parse_duration() helper. All four subcommands (session, weekly, weekly-sonnet, monthly) expose the flag.
• app.py: CCBurnApp.__init__() accepts and stores recent_window_minutes.

Screenshots

Before / after screenshot

Test plan

• pytest — 105 tests pass (29 new tests covering window filtering, insufficient-points fallback, backward compatibility, and span-check scaling)
• ruff check — no issues
• Manual test: idle 2h + active 30min scenario shows steeper projection with --recent-window 30m