counter — atomic distributed counters (in-RAM, no DB)

Monotonic build numbers / ids shared across machines. GET a name to get the
NEXT integer — the GET MUTATES (it increments and hands you the new value),
so two machines racing the same name never collide.

NEXT NUMBER (this GET increments — it is NOT a pure read):
  curl https://fleet-counter.0exec.com/ci-build            # → 1   then 2, 3, ... on each call, from anywhere

PEEK (read the current value WITHOUT incrementing; 0 if it never existed):
  curl 'https://fleet-counter.0exec.com/ci-build?peek'     # → 3

BUMP / SET / RESET / DELETE (POST):
  curl -X POST 'https://fleet-counter.0exec.com/ci-build?by=10'    # += 10 (N may be negative) → new value
  curl -X POST 'https://fleet-counter.0exec.com/ci-build?by=-1'    # -= 1
  curl -X POST 'https://fleet-counter.0exec.com/ci-build?set=100'  # set to 100 → 100
  curl -X POST 'https://fleet-counter.0exec.com/ci-build?reset'    # set to 0 → 0
  curl -X DELETE  https://fleet-counter.0exec.com/ci-build         # remove the counter (204)

REAL USES:
  N=$(curl -s https://fleet-counter.0exec.com/ci-build)            # grab the next CI build number
  ID=$(curl -s https://fleet-counter.0exec.com/order-seq)          # monotonic id with no shared DB
  docker build -t app:$(curl -s https://fleet-counter.0exec.com/img-tag) .

Notes:
  * Names: [A-Za-z0-9._-], 1-128 chars, must start alnum.
  * Values are signed 64-bit. An increment that would overflow int64 is refused
    with 409 (never silently wrapped) — use ?set to rebase. In RAM only — a
    restart resets every counter to 0, so these are best-effort monotonic ids,
    not a durable ledger.
  * Counters idle past the TTL (default 24h) are evicted; a fresh GET restarts
    them at 1. There's a hard cap on distinct counters → 429 when full.
  * Writes are per-IP rate-limited; ?peek reads are not.
  * zsh "no matches found"? Quote any URL with a ?  e.g. 'https://fleet-counter.0exec.com/x?peek'.