add simulated 5h/7d usage limits via [rate_limits] config

[tmck-code]

Context

Claude Code only supplies rate_limits in its stdin JSON on some setups. Without it, YAS rendered
the 5h bucket as ∞ and omitted the 7d bucket entirely, so users on those setups had no way to see
or reason about their usage windows.

This adds an opt-in simulator so a setup with no real limit data can model the 5h/7d windows
locally, and makes the two buckets behave consistently when idle.

Ships in v0.9.0.

Changes

New [rate_limits] config table

• Inline five_hour / seven_day tables with budget, window, anchor and epoch:

  [rate_limits]
  five_hour = { budget = 44_000_000, window = "5h", anchor = "rolling" }
  seven_day = { budget = 440_000_000, window = "7d", anchor = "fixed", epoch = "0 0 * * 0" }

• Defining a key synthesises that bucket and overrides any real payload value; omitting it
  leaves existing behaviour untouched.

• used_percentage is derived from trailing-window TickRecord history against budget, clamped
  to 0-100.

• resets_at is a genuine future timestamp, so the burndown trend isn't suppressed.

• anchor = "rolling" runs now-window .. now; anchor = "fixed" aligns to the cron epoch.

Cron subset (claude/yas/cron.py)

• Stdlib-only 5-field parser supporting *, integers, comma lists and */n steps.
• Anything outside that subset is a load-time error rather than a silent mis-anchor.
• Load-time errors also for epoch with anchor="rolling", and anchor="fixed" without epoch.
• One bad bucket doesn't block the other.

Minute-granularity updates

• Derived values are quantized to the minute: stable within a minute, recomputed on rollover.
• A named regression guard pins the interval at 60s so it can't drift onto the shared 300s
  CACHE_TTL_SECONDS.
• An in-process bucket cache additionally dedupes usage_since scans within mon's polling loop.
  (It cannot span statusline renders, since each render is a fresh process — the cross-render
  stability comes from the quantization, not the cache.)

Consistency fix

• An idle 7d bucket now renders ∞ like 5h, instead of being omitted. This is a change to default
  behaviour, independent of the new config.

Checklist

• Tests added or updated for new/changed behaviour
• (bug fixes) a regression test now covers the situation so it can't come back
• N/A — no behaviour change, because:

Screenshots / recording

	before / after
kitchen-sink
idle-7d
rate-limits-config-synthesized

kitchen-sink is the default demo fixture; idle-7d is a hand-crafted variant (no existing demo
scenario has an idle seven_day bucket — all 50 set a nonzero seven_day_pct) confirming that
main omits the idle 7d bucket while this branch renders ∞. The new cell picks up a matching
┬/┴ divider pair and the right box edge stays aligned; the full demo gate over all 50 scenarios
showed border/elbow glyphs byte-identical (the only diffs were pre-existing ±1-cell sparkline
jitter that reproduces on main).

rate-limits-config-synthesized demonstrates this PR's headline feature end-to-end: stdin supplies
no rate_limits payload at all, but a [rate_limits] config table is set. Before (main) has
no config support and no stdin data, so it shows the exact bug described in Context (5h ∞, 7d
omitted). After (this branch), the config-driven simulator synthesizes real usage from local
history — 5h ≈ 53.3%, 7d ≈ 36.9% — instead of ∞/omitted.

• N/A — no visible change

Benchmark

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
main	45.4 ± 1.7	42.3	50.6	1.00
PR	46.3 ± 2.9	43.1	63.5	1.02 ± 0.07

Within noise — the error bars overlap.

• N/A — no performance-relevant change

System info

Key	Value
OS	Linux debian-work 6.12.105+deb13-amd64 #1 SMP PREEMPT_DYNAMIC Debian 6.12.105-1 (2026-08-24) x86_64 GNU/Linux
Claude Code	2.1.250 (Claude Code)
Terminal	TERM=tmux-256color TERM_PROGRAM=tmux SHELL=/bin/bash COLORTERM=truecolor
Locale	LANG=en_AU.UTF-8 LC_ALL=
Python	Python 3.12.7
uv	uv 0.9.17

🤖 Generated with Claude Code