feat: add DakeraMemoryStore (Strands MemoryStore integration)

[ferhimedamine]

Problem

strands-dakera exposed Dakera only as the dakera_memory tool — the model has to call it explicitly. Strands ≥ 1.45 ships a strands.memory module with a MemoryStore extension point that a MemoryManager drives inside the agent loop: it searches stores to recall context (injected into the prompt automatically) and, when writable, writes new memories — directly or via periodic extraction from the conversation. This adds that second, out-of-the-box integration.

Design

• DakeraMemoryStore(MemoryStore) (store.py) implements the two methods that matter for a vector-DB-style backend:
  • search(query, options) → Dakera decay-weighted recall, mapped to MemoryEntry (server-side id/score/importance/created_at preserved in metadata). max_search_results precedence: per-call SearchOptions → configured default → 5.
  • add(content, metadata) → a single client-side write sink (store_memory). Because the store implements add (not add_messages), enabling extraction uses the manager's client-side ModelExtractor.
  • add_messages / initialize / get_tools are intentionally left as the inherited Protocol stubs so Strands' _has_method detection reports them as absent (verified in tests).
• Reuses the existing DakeraServiceClient; the client is built lazily on first use, and now accepts optional base_url / api_key (env fallback unchanged — backward compatible).
• strands-agents floor bumped >=1.0.0 → >=1.45.0 (first release exposing strands.memory).

Usage

from strands import Agent
from strands.memory import MemoryManager
from strands_dakera import DakeraMemoryStore

store = DakeraMemoryStore(agent_id="alex", writable=True, extraction=True)
agent = Agent(memory_manager=MemoryManager(stores=[store]))

agent("Remember that I prefer dark-mode dashboards.")
agent("How do I like my dashboards?")  # recalls the stored preference

The dakera_memory tool is unchanged and can be used alongside the store against the same server.

Testing

• tests/test_store.py — 12 tests with a mocked DakeraServiceClient (no live server/SDK): protocol conformance, write-sink detection via _has_method/_has_write_sink, MemoryEntry mapping, top_k precedence, and lazy client construction.
• Full suite: 28 passed (16 existing tool + 12 store). ruff format --check, ruff check, and mypy src all clean — verified against real strands-agents 1.45.

Checklist

• hatch run format / ruff check clean
• mypy src clean
• Tests pass on the mocked client (py3.10–3.13 matrix in CI)
• README + python/README.md updated
• Backward compatible (tool API and DakeraServiceClient() unchanged)

Context: requested by a Strands maintainer on the community docs PR (strands-agents/harness-sdk#3082) — "would you be able to integrate as a store instead?"