test: add unit + integration test infrastructure for kernel components

[RyamL1221]

Summary

Adds CI test infrastructure covering components that previously had zero automated coverage, and wires up two new GitHub Actions workflows to run them.

New workflows:

• test-unit.yml — fast (~2-3 min), no external services, runs on every PR/push to main
• test-memory-integration.yml — spins up Ollama + ChromaDB, runs real Mem0 provider and cross-agent shared-memory integration tests (~10-15 min)

New test coverage (123 tests):

• MemoryFormatter — 27 tests, all formatting branches (profile/task_context/conversation, generic fallback, malformed input, exception path)
• IntentRouter — 35 tests, keyword classification scoring, confidence thresholds, LLM fallback (mocked, no network calls)
• ConversationExtractor — 12 tests, metadata construction, user_id fallback, error handling
• FIFOScheduler — 17 tests, _execute_syscall and _execute_batch_syscalls execution paths, including failure handling
• SyscallExecutor — 21 tests, request dispatch, syscall factory, write-barrier stamping (mocked against real MemoryWriteBarrier via spec=)
• FastAPI endpoints (launch.py) — 11 tests, request validation, /status, /query, /user/select routing, all with mocked kernel components (no real LLM/memory backend, verified with all provider API keys unset)

New integration test:

• test_shared_memory_e2e.py — full write → retrieve → inject → format pipeline across two writer agents and one reader agent, against real Ollama + ChromaDB. Skips gracefully (pytest.skip) if Ollama is unreachable, verified via a test-only OLLAMA_SKIP env var that exercises the real reachability-check code path.

Infrastructure:

• pytest.ini + tests/conftest.py register an integration marker so pytest -m "not integration" / -m integration cleanly separate the two test tiers
• scripts/patch_config_for_memory_tests.py — small script to configure config.yaml for the mem0/Ollama/ChromaDB test setup, reused by both the workflow and the E2E test
• README updated with a "Running Tests" section: how to run unit vs. integration tests, per-file skip-vs-fail behavior for integration tests, and a table of known pre-existing issues (see below)

Known pre-existing issues (found, not introduced, by this PR)

These are excluded via --ignore in test-unit.yml and documented in the README so they don't block CI or confuse future contributors:

File	Issue
tests/modules/llm/openai/test_concurrent.py	Broken import (pre-existing)
tests/modules/llm/openai/test_single.py	Broken import (pre-existing)
tests/modules/memory/test_memory.py	Broken import (pre-existing)
tests/modules/memory/test_memory_user_identity_isolation.py	Order-dependent failure — passes in isolation, fails in full suite due to ConfigManager singleton state pollution from earlier tests

None of these are caused by this PR; they were discovered while wiring up the unit workflow and are called out explicitly rather than silently worked around.

Testing

• All new unit tests verified locally, passing: 375 total unit tests (252 pre-existing + 123 new), 0 failures
• test_shared_memory_e2e.py verified locally against real Ollama + ChromaDB (pass) and with OLLAMA_SKIP=1 (skip)
• All workflow YAML files validated (yaml.safe_load)
• cancel-workflow.yml confirmed to already cover the two new workflows (workflow_id: all, no update needed)
• This is the first run of test-unit.yml / test-memory-integration.yml against GitHub's actual runners — opening this PR specifically to validate them in the real CI environment before merge

Documentation

No docs/ changes — this PR adds test infrastructure and coverage only, no new features or behavior changes to document.