feat: add LM Studio as first-class LLM provider

[ramirolaso]

Summary

• Adds LlmProvider::LmStudio variant alongside the existing Claude, Ollama, and Custom providers
• Injects ANTHROPIC_BASE_URL=http://10.0.2.2:1234 and ANTHROPIC_API_KEY=lm-studio into the guest VM so claude-code talks to the local LM Studio server via the SLIRP gateway — no proxy needed (LM Studio 0.3.x+ exposes a native
  Anthropic-compatible Messages API)
• Adds lm_studio(model) and lm_studio_with_host(model, host) constructors, and extends all match arms: cli_args(), env_vars(), is_local(), description(), and the model() builder
• Adds LM_STUDIO_MODEL detection in detect_llm_provider() (checked before OLLAMA_MODEL)
• Adds examples/lm_studio_local.rs — a runnable example mirroring ollama_local.rs, with KVM/mock fallback and LM Studio-specific prerequisites documented
• Bumps default VM memory to 2048 MB across all local-LLM examples and common::make_box — necessary because claude-code (~250 MB RSS) + a full initramfs (~299 MB decompressed) exceeds 1 GB

Test plan

• cargo test -p void-box -- llm::tests — all 17 tests pass (5 new LM Studio tests added)
• cargo build — compiles clean
• cargo clippy -- -D warnings — no warnings
• Mock mode: LM_STUDIO_MODEL=deepseek-r1-distill-llama-8b cargo run --example lm_studio_local runs without KVM
• KVM mode: example boots guest, vsock connects, and claude-code calls LM Studio at http://10.0.2.2:1234

LM Studio prerequisite

In the LM Studio app → Local Server tab → Start Server (default port 1234). The server must listen on 0.0.0.0, not just 127.0.0.1, for the SLIRP gateway to reach it from inside the VM — same constraint as Ollama.

[Copilot]

Pull request overview

This PR adds LM Studio as a first-class LLM provider alongside the existing Claude, Ollama, and Custom providers. It enables void-box to use local LM Studio instances running on the host machine, leveraging LM Studio 0.3.x+'s native Anthropic-compatible API via SLIRP networking. The implementation also includes a refactoring of the kernel module installation script to support newer .ko.zst compression formats, and bumps VM memory for local LLM examples to accommodate the larger resource requirements.

Changes:

• Adds LlmProvider::LmStudio variant with constructor methods and full integration into the provider system
• Refactors kernel module installation in build script to support .ko.xz, .ko.zst, and uncompressed .ko formats
• Adds comprehensive test coverage for LM Studio provider (5 new tests)
• Increases VM memory from 1024 MB to 2048 MB for local LLM examples to handle claude-code + initramfs requirements
• Adds LM_STUDIO_MODEL detection to the provider auto-detection logic

Reviewed changes

Copilot reviewed 5 out of 6 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
src/llm.rs	Adds LmStudio enum variant, constructors, and implementations for all provider methods (cli_args, env_vars, is_local, description, model builder); includes comprehensive test coverage
scripts/build_guest_image.sh	Refactors kernel module installation to handle multiple compression formats (.ko.xz, .ko.zst, .ko) with built-in module detection
examples/lm_studio_local.rs	New runnable example demonstrating LM Studio integration with KVM/mock fallback, mirroring the ollama_local.rs pattern
examples/ollama_local.rs	Updates memory allocation from 256 MB to 2048 MB for local LLM workloads
examples/common/mod.rs	Adds LM_STUDIO_MODEL detection in detect_llm_provider() and updates default memory to 2048 MB
.gitignore	Adds IDE and local utility directories (.claude, .idea, .local-utils)

---

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

[Copilot]

The documentation comment should be updated to include the newly added LM Studio detection. It currently lists Ollama and LLM_BASE_URL options, but omits LM_STUDIO_MODEL which is now checked first in the implementation.

[Copilot]

The documentation references scripts/build_claude_rootfs.sh, but the Ollama example uses scripts/build_guest_image.sh. These should be consistent. Verify which script name is correct and update the documentation accordingly.

Suggested change

	//! scripts/build_claude_rootfs.sh
	//! scripts/build_guest_image.sh