feat(memory-core): add L1 atomic memory import API with dedup

[elementlo]

Add a user-driven L1 atomic memory import path that bypasses the
L0->L1 distillation pipeline, letting users bulk-write personal
memory records (from markdown/json/txt files or JSON arrays) directly
into the L1 atomic layer for model recall.

MemoryCore kernel

• v2-router.ts: add handleAtomicCreate handler, register /atomic/create
  route (dual-mounted on /v2 and /v3), batch create up to 500 records
  per request with server-generated record_id (m_{timestamp}_{hex})
• Dedup on create: before writing, search existing L1 memories with a
  3-tier fallback — vector search (searchL1Vector) -> FTS
  (searchL1Fts) -> exact content match (queryL1Records, works in
  standalone mode without embedding/FTS5). Modes: skip / update,
  configurable threshold (default 0.85). Isolated by team+agent+user.
• generated/schemas.ts + types.ts: atomicCreateRequestSchema /
  atomicCreateDataSchema and AtomicCreateRequest / AtomicCreateData
  types (with dedup field and skipped/updated response counters)
• v2-schemas.ts: re-export the new schemas and types

MemoryPanel

• chat-memory.ts: POST /chat-memory/import-memories route accepting a
  JSON array or file text (.md/.json/.txt), parsed then forwarded to
  /v3/atomic/create
• chat-memory-import-parser.ts (new): markdown/JSON/TXT parser, splits
  by ## heading or paragraph
• web: standalone ImportMemories page as a left-menu item (parallel to
  Chat Memory, not a tab inside it) — ImportMemoriesPanel supports
  paste/upload, live parse preview, optional dedup, and lists existing
  L1 memories after import

SDK

• TypeScript: createMemories method + V3AtomicCreateRequest/Record/
  Dedup/Data types
• Python: create_memories method (sync + async)

Design decisions

• Isolation inherits the V3 router's team+agent+user triple check
• Soft classification via type + scene_name (aligned with existing L1
  model); no URI namespace
• handleAtomicUpdate untouched (keeps update-only semantics; create
  goes through the new endpoint)

Tested

• skip mode: identical content -> created=0, skipped=1
• update mode: identical content -> created=1, updated=1, id reused
• distinct content -> created=1
• user isolation: other-user cannot query test-user's memories
• MemoryCore typecheck 0 errors; tsx boots on 8420

[Maxwell-Code07]

Thank you so much for your attention and contribution! We will arrange an internal review for this PR shortly, and all feedback will be shared right here in the discussion.