feat: mature lldb agent harness

[AiMiDi]

Description

Matures the existing LLDB harness into the v1.0 AI debugging adapter layer.

This PR adds a formal stdio Debug Adapter Protocol entry point (cli-anything-lldb-dap) while keeping the existing JSON CLI and REPL workflows intact. It hardens the persistent CLI session daemon, makes unresolved breakpoints honest by default, expands stop/process/source/disassembly metadata, and updates the agent-facing docs for DAP and persistent-session workflows.

Follow-up GUI/debugger responsiveness improvements are included: DAP continue now returns before blocking LLDB continue completes, pause uses async interrupt, and breakpoint mutation during active continue first interrupts and waits for a stopped state. launch/attach support the compatibility autoContinueInternalBreakpoints flag for known debugger-internal JIT/startup stops.

Latest iteration adds structured DAP stop rules and external profile files. Targets such as C4D can inject rules matching reason, module, function, and/or regex; rules can either classify a stop or auto-continue it. DAP stopped events now expose body.cliAnythingStop.origin (manualPause, internalTrap, or debuggee) with matched-rule and frame/module/function metadata, so AI clients can distinguish a requested pause from NVIDIA/C4D/internal traps. Existing running cli-anything-lldb-dap processes do not hot-load profile/code changes; clients must restart the adapter and re-attach/re-launch.

Skip/testing note: the current local Windows run has 0 skipped tests. Earlier local runs could skip the optional core-load negative-path scenario when LLDB_TEST_CORE was unset; after rebasing onto current main, that test now uses a placeholder fixture, so no skip waiver is needed in this PR.

Type of Change

• New Feature — adds new functionality to an existing harness or the plugin
• Bug Fix — fixes incorrect behavior
• Documentation — updates docs only

Checklist

• Code follows existing patterns and conventions
• Existing JSON CLI / REPL workflows remain intact
• DAP stdout remains protocol-only; diagnostics go to stderr or log files
• Commit messages follow the conventional format (feat:, fix:, docs:, test:)
• I have tested my changes locally

Test Results

# Windows local environment, from lldb/agent-harness
$ python -m py_compile cli_anything\lldb\dap.py cli_anything\lldb\core\session.py cli_anything\lldb\lldb_cli.py
passed

$ python -m pytest cli_anything\lldb\tests\test_core.py -q
46 passed in 0.30s

$ python -m pytest cli_anything\lldb\tests -q -rs
53 passed, 2 warnings in 40.36s
Warnings are LLDB SWIG DeprecationWarning messages from the local LLDB Python bindings.
Skip status: 0 skipped.

# Repository-level skill mirror check
$ python .github\scripts\validate_root_skills.py
Root skills validation passed.

$ git diff --check
passed, with Windows CRLF normalization warnings only

[yuh-yang]

Thanks for the Apr 27/28 updates. The DAP responsiveness and stop-rule additions are useful.

I checked the current head (f773006) and the attach issue still looks open: _handle_attach() still requires program/executable before it accepts pid/processId. Common DAP attach configs often provide only a process id, and LLDB can attach that way, so please support pid-only attach or document the stricter schema clearly.

Since the PR is still draft, I would also keep the real-client smoke checklist before marking ready: initialize, attach/launch, breakpoints, configurationDone, threads, stackTrace, variables, continue, and disconnect.