fix: gzip torch profiler traces by default

[CUHKSZzxy]

Motivation

PyTorch profiler traces can grow large, while LMDeploy currently exports them as plain JSON. vLLM enables gzip by default for the same export_chrome_trace flow to reduce storage use.

Changes

• Export LMDeploy PyTorch profiler traces as .json.gz by default.
• Allow uncompressed JSON with LMDEPLOY_PROFILE_USE_GZIP=0.
• Document the setting in the English and Chinese profiling guides.

Validation

• Verified that the default output is valid gzip-compressed JSON.
• A profiler smoke capture decreased from 49,729 bytes to 4,508 bytes (90.9% smaller).
• Verified that the opt-out writes valid uncompressed JSON.
• File-level pre-commit hooks passed.

Assistance

Assisted with Codex + GPT-5.5 xHigh Fast, reviewed manually

[Copilot]

Pull request overview

This PR reduces the storage footprint of LMDeploy’s PyTorch profiler Chrome trace exports by switching the default trace output to gzip-compressed JSON, while keeping an environment-variable opt-out for plain JSON.

Changes:

• Add LMDEPLOY_PROFILE_USE_GZIP (default True) to control whether profiler traces are written as .json.gz or .json.
• Update the profiler dump path logic to emit .json.gz by default.
• Document the new setting in both English and Chinese profiling guides.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
lmdeploy/pytorch/envs.py	Introduces torch_profile_use_gzip env-backed flag (LMDEPLOY_PROFILE_USE_GZIP, default on).
lmdeploy/pytorch/engine/model_agent/profiler.py	Switches trace output suffix to .json.gz when gzip is enabled.
docs/zh_cn/advance/pytorch_profiling.md	Documents default gzip behavior and the opt-out env var.
docs/en/advance/pytorch_profiling.md	Documents default gzip behavior and the opt-out env var.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.