Refactor: replace level magic numbers with LevelPolicy object #83

Problem

The level 1 (project) redesign in #80 introduced == 1 / > 1 / < 2 / {0, 1} magic number comparisons scattered across 6 files:

• server.py: _resolve_level, store_knowledge, negate_knowledge, delete_knowledge
• sqlite.py: _build_filter
• cli.py: _build_hierarchy_filters, _cmd_config_add_level
• manager.py: get_project_config validation
• engine.py: _sync_project_levels

Adding a third storage behavior (e.g., level 0 writes to a local file) would require touching all of these.

Proposed solution

Replace the raw level integer with a LevelPolicy dataclass returned by _resolve_level:

@dataclass
class LevelPolicy:
    level: int
    name: str
    repo_url: str | None
    repo_branch: str | None
    writable: bool
    stores_locally: bool      # level 0 and 1: True, level 2+: False
    creates_pr: bool           # level 1 and 2+: True, level 0: False
    locally_deletable: bool    # level 0 and 1: True, level 2+: False
    pr_path_fn: Callable[[str], str] | None  # level 1: .lore/knowledge/<key>.md, level 2+: <key>.md

Then store_knowledge becomes:

policy = _resolve_level(level)
if policy.creates_pr and policy.repo_url:
    pr_url = _create_shared_pr(..., file_path_override=policy.pr_path_fn(key) if policy.pr_path_fn else None)
if policy.stores_locally:
    # SQLite upsert

No branching on magic numbers. New level behaviors are added by defining a new policy, not by editing every consumer.

Files affected

• src/lore/mcp/server.py — _resolve_level returns LevelPolicy, consumers use policy fields
• src/lore/store/sqlite.py — _build_filter takes implicit_levels param instead of hardcoding {0, 1}
• src/lore/config/models.py — LevelPolicy dataclass
• src/lore/cli.py — _build_hierarchy_filters uses policy
• src/lore/sync/engine.py — _sync_project_levels uses policy

Origin

Identified by /simplify altitude review on #80 changes.