docs: update CLAUDE.md to instruct AI to run install.sh and TestFromTestdata locally

[fennoai]

Problem

When AI assistants (like Claude Code / xgopilot) modify llcppg components (e.g., _xtool/internal/parser), they currently mark TestFromTestdata tests as "requires llcppsigfetch tool via CI" in PR descriptions:

- [ ] TestFromTestdata tests pass (requires llcppsigfetch tool via CI)

This is incorrect — the AI assistant can and should install these tools locally by running bash ./install.sh, then run the tests to verify changes.

Requested Change

Update CLAUDE.md to explicitly instruct AI assistants that:

1. After modifying llcppg components (especially _xtool/ code or anything affecting TestFromTestdata), they should run bash ./install.sh to install/rebuild llcppsigfetch, llcppsymg, and other LLGo-compiled tools locally.
2. Then run TestFromTestdata tests via go test -v ./cl/internal/convert -run TestFromTestdata to verify changes, rather than deferring to CI.
3. Update the expected test data if the output format changes — use gen:true to regenerate gogensig.expect files, verify correctness, then switch back to gen:false.

Suggested CLAUDE.md Addition

Add a section like:

### After Modifying LLGo-Compiled Components

When you modify code under `_xtool/` (e.g., `_xtool/internal/parser`), you MUST:

1. Run `bash ./install.sh` to rebuild and install the LLGo-compiled tools (`llcppsigfetch`, `llcppsymg`, etc.)
2. Run `go test -v ./cl/internal/convert -run TestFromTestdata` to verify the changes
3. If test output changes are expected, temporarily set `gen:true` in the test to regenerate expected output, verify it, then set back to `gen:false`
4. Do NOT skip these tests or mark them as "requires CI" — they can and should be run locally after `install.sh`

Context

Raised by @luoliwoshang in PR #637.

[luoliwoshang]

/code

[fennoai]

CLAUDE.md updated with install.sh and TestFromTestdata instructions

Added a new "After Modifying LLGo-Compiled Components" section to CLAUDE.md under "Building and Testing" that instructs AI assistants to:

1. Run bash ./install.sh to rebuild LLGo-compiled tools after modifying _xtool/ code
2. Run go test -v ./cl/internal/convert -run TestFromTestdata to verify changes locally
3. Use gen:true to regenerate expected output when needed, then switch back to gen:false
4. Never skip these tests or mark them as "requires CI"

PR: #639