deps: bump processkit =1.1.0 → =1.2.0 (deep-audit fixes)

[ZelAnton]

Bumps the pinned processkit crate =1.1.0 → =1.2.0 (published 2026-07-03, a deep-audit release). Non-breaking within 1.x — the binding compiles with zero code changes and all tests pass.

What this brings

Almost entirely inherited correctness fixes on surface processkit-py exposes:

• Command timeout now enforced while streaming a shared-group handle (group.astart(cmd) → stdout_lines) — previously the deadline watchdog armed only for own-group handles, so a quiet child left the stream pending forever. (New regression test.)
• Reusing a ProcessGroup after a graceful shutdown(escalate_to_kill=False) no longer orphans fresh children.
• wait/profile gain a 64 MiB in-flight cap (OOM guard); bounded ProcessResult Debug; cleaner Error::Timeout Display; cassette replay fidelity + crash-safe writes; env-isolation no longer pierced by client defaults; ok_codes([]) a true no-op; Windows suspend/resume thread-owner check; cgroup hybrid-mount detection.

The one upstream behavioral change (default_env duplicate-key → last-wins) does not affect this binding (CliClient env is a dict).

Deferred (surface stays minimal for the 1.0 freeze)

The new 1.2.0 additive APIs — no_timeout, output_contains_any, default_retry/RetryPolicy, default_env_fn (and the Rust-ergonomic Box/Arc<dyn ProcessRunner> + Error accessors, N/A to the Python surface) — are deferred. They're safely additive after the freeze.

Verification

• cargo build / clippy -D warnings / fmt clean against 1.2.0; Cargo.lock adds zero new transitive deps.
• ruff, mypy --strict, mypy.stubtest, pytest 205 passed / 4 skipped (+1: test_shared_group_streaming_enforces_command_timeout).