Implement offline ATAK assistant plugin with local marker create/edit ops

[nbschultz97]

Motivation

• Provide an offline-first ATAK-facing assistant that can answer queries locally and perform constrained, auditable local operations against a nearby ATAK/CoT server.
• Enable safe natural-language-driven workflows for marker management (create/edit) so operators can quickly annotate area information from the app without cloud dependence.
• Keep the design minimal and modular to preserve future integration with CSI/pose detection features and other edge ML components.

Description

• Replace the TAK plugin stub with a bound service TakPluginService that exposes ask(...) and runs an offline LLM path plus optional local operation execution via a planner/executor.
• Add TakOperationPlanner to parse constrained commands (create marker uid=... lat=... lon=... type=... and edit marker ...) and a TakOperationExecutor which keeps a local marker cache and publishes CoT via existing core interop APIs.
• Wire up the app UI by updating LocalChatActivity and activity_local_chat.xml to bind to the plugin service, send prompts, and display assistant replies.
• Provide an improved offline deterministic LocalLlm implementation that documents supported command formats and echoes prompts for local guidance, and update tak-plugin build config to depend on :core and include unit test support.

Testing

• Added unit tests TakOperationPlannerTest to validate parsing of create marker and non-action prompts, which are included under tak-plugin/src/test.
• Attempted to run gradle :llm:test :tak-plugin:testDebugUnitTest, but the CI/container environment failed because the Android SDK is not configured (missing ANDROID_HOME / local.properties).
• The tests should pass in a properly provisioned environment; to run locally use ./gradlew :llm:test :tak-plugin:testDebugUnitTest or gradle :llm:test :tak-plugin:testDebugUnitTest after setting up the Android SDK.

---

Codex Task