feat(chat/synapse): multi-turn memory (#129), hub-spoke delegation (#262), and agent cards (#272)

[sbharatjoshi]

This PR implements the ConversationReflector and ConversationSummarizer integration, resolving multi-turn chat history bugs, porting the companion prompt, resolving build flakiness in KeyedSynapticTagEncoderTest, implementing the multi-agent coordinator hub-spoke delegation, and exposing Agent Cards capability discovery manifests and REST endpoints.

Changes Included:

1. Multi-Turn Session Integration (#129)

• Companion Prompt: Ported companion-system.txt prompt template from spector-enterprise to spector core repository (spector-synapse module resources).
• Graph Multi-Turn Seeding: Updated AgenticChatGraph.java to support an overloaded chat method that accepts conversation history as a list of ChatMessage objects, prepending it to the state graph's messages channel.
• ChatService Wiring: Updated ChatService.java to load session history (supporting both database loading and client-passed arrays), run compaction via ConversationSummarizer when exceeding the token budget, map history to LangChain4j message types, and execute the graph with the fully reconstructed history.
• Test Hardening: Fixed a bug in the integration test in-memory mock InMemoryChatMemoryPort in ChatServiceLiveIT.java and DeepConversationLiveIT.java where excludeSessionId was ignored (which was masking bugs by leaking active session turns as cross-session semantic recalls).
• Flaky Test Resolution: Resolved build flakiness in KeyedSynapticTagEncoderTest.differentKeysProduceDifferentFilters where two different keys could randomly collide in the 64-bit Bloom filter space for a single tag. It now asserts over multiple tags to ensure a 0% collision probability.

2. Multi-Agent Hub-Spoke Delegation (#262)

• AgentSelector: Created capability-based agent selector component that matches sub-tasks to the best available child agent by matching keyword and capability heuristics.
• AgentDelegationNode: Created node action that executes specialized child agents via AgenticChatGraph.chat inside the compiled plan subgraph.
• State Channels: Upgraded both CognitiveState and CoordinatorState schemas with child_results appender channels.
• Dynamic Compilation: Updated DynamicGraphBuilder to use the AgentSelector and wire AgentDelegationNode when compiling AGENT nodes in subgraphs.
• Planner Priming: Modified PlannerNode and coordinator-planner-system.txt to automatically list and inject all available child agent metadata into the planner's context prompt.
• Result Aggregation: Updated SubgraphExecutorNode to fetch child execution results from completed subgraphs and bubble them up to the coordinator state.
• CoordinatorGraphTest: Added a comprehensive integration test verifying the entire planner-executor-evaluator lifecycle and child results propagation.

3. Agent Cards: JSON Capability Discovery Manifest (#272)

• AgentCard: Created the capability discovery manifest record model.
• REST Endpoints: Exposed GET /api/v1/agents/soul/card, GET /api/v1/agents/{id}/card, and POST /api/v1/agents/{id}/invoke in AgentController.
• Well-Known Discovery: Created WellKnownController to expose the active default agent card at the standard A2A path GET /.well-known/agent.json.
• AgentCardApiTest: Added integration test suite verifying the REST API mappings and JSON structure outputs.

Resolves #129
Resolves #262
Resolves #272

[github-actions]

All contributors have signed the CLA. ✅
_{Posted by the CLA Assistant Lite bot.}

[sbharatjoshi]

I have read the CLA Document and I hereby sign the CLA

[sbharatjoshi]

recheck

[jarvispectrayan]

I have read the CLA Document and I hereby sign the CLA

[jarvispectrayan]

recheck