Intermittent hang at 'booting location services': silent idle, no CPU/net; headless repro (resume existing session); references #35870/#41934

Environment

• opencode 1.18.30 (latest at 2026-09-12), Linux x86_64, latest channel
• Effect @effect/... 4.0.0-beta.83 (per the root cause described in #35870)

Symptom

Startup intermittently hangs at the "booting location services" phase:

1. Phase 1 (load/config/init) completes fast (~2-5s, all LSPs / init count logged).
2. booting location services is logged, then the log goes completely silent and the
   process idles: main thread in epoll_wait/cond-wait, 0% CPU, no disk, no sockets.
3. The session is never created (created id=... never logs). Hangs I captured lasted
   150s, 173s, 1519s and "forever" (killed by watchdog).

It is intermittent: the same command in the same directory alternates between completing
the location boot in under a second and hanging for minutes.

Empirical isolations (all reproduced headlessly, no TUI involved)

• Reproduces in a fresh scratch dir with an empty opencode.json (no plugins, no
  project config) => not plugin/provider/cwd related. opencode run -c ... (fresh session)
  and opencode run --session <existing-session> -c ... are both affected.
• With HTTP_PROXY/HTTPS_PROXY/ALL_PROXY pointed at a 502-returning logging CONNECT proxy,
  the boot still hangs with ZERO sockets open => the stall is a local effect await,
  not a network egress issue.
• SQLite (memory.db) IMMEDIATE lock acquired instantly during a hang => not a DB lock.
• 16/16 cold boots completed (3-23s) under pure CPU load (4x yes), but hangs are
  provoked by parallel cold boots (2-4 concurrent opencode instances / the TUI
  booting an existing project dir eagerly) - matching #35870's "scheduler/memory pressure".
• opencode serve + POST /session boots lazily and creates a session in ~5s without
  hanging; the interactive TUI (and run --session) creates/attaches its location eagerly
  at startup, which is the path that strands.

Repro

# 1) create a session quickly
opencode run --dir /tmp/ocr-repro -c "hi" --print-logs   # fast; note the session id
# 2) resume it (this routes through the eager location boot that intermittently hangs)
for i in $(seq 1 10); do
  opencode run --dir /tmp/ocr-repro --session <id> -c "hi" --print-logs &
done

A fraction of the instances stall right after booting location services with a silent
log and no session created; they must be killed. Kill + retry makes the same command
complete fine.

References

• #35870 (same root cause: InstanceStore.load boot Deferred + effect beta.83 runLoop
  re-entrancy guard -> lost wakeup). This issue adds a headless repro + "not network /
  not plugins / not DB" isolation and the --session eager-location-boot trigger.
• #41934 (unguarded Npm.install/arborist reify also stalls startup when the registry
  is slow; observed as a separate first-phase delay of 12-120s on the npm-plugin config).