Refactor SandboxThread init/reset state setup helpers

[seanwevans]

Motivation

• Reduce duplicated initialization logic by centralizing immutable/config wiring and transient runtime state so __init__ and reset share identical setup behavior.
• Make capabilities deserialization, quota/policy wiring, NUMA/cgroup pointers and transient counters initialization explicit and easier to maintain.
• Make lifecycle expectations for attach-message deduplication explicit so message-id behavior is intentional and auditable.

Description

• Add a private _init_config_wiring(...) helper to centralize configuration/immutable wiring including quotas, merged imports, NUMA/cgroup pointers, and capabilities deserialization.
• Add a private _reset_runtime_state() helper to centralize resetting transient counters and runtime fields such as _cpu_time, _mem_peak, _ops, _errors, _latency, _latency_sum, _trace_enabled, _syscall_log, _start_time, _open_files, _network_ops, _output_bytes, and _child_work plus quarantine/termination fields.
• Update __init__ and reset to call the two helpers so both construction and reuse flows share the same initialization behavior while keeping existing public method signatures and behavior unchanged.
• Add an inline comment near _next_attach_msg_id / _seen_attach_msg_ids clarifying that the dedup set spans sandbox lifetimes and that attach message IDs are monotonic and intentionally preserved across reset() to avoid stale replay collisions.

Testing

• Ran python -m compileall -q pyisolate/runtime/thread.py which completed successfully.
• Ran full test suite with python -m pytest -q; the run produced repository-wide failures (106 failed, 101 passed, 1 error) caused by a mismatch in the test stub (_StubBPFManager.load() receiving an unexpected mode keyword) and are unrelated to this refactor.

---

Codex Task