[pmeems]

Phase 1 of the Map Model and Basic Navigation plan.

Motivation

Core had no notion of a map. ActiveX holds one dataset and its extent directly. Later phases (viewport, multi-layer rendering, navigation, drag-and-drop) need one authoritative Core model that owns the layers and their combined extent.

What changes

• mwg::Map / mwg::MapLayer (mapwingis/map/Map.h): an ordered layer collection with add, remove by identity, clear, count, position lookup, find, and visibility.
• Map::GetExtent(ExtentScope): the union of valid layer extents. VisibleLayers (the default, used for zoom-to-map) excludes hidden layers; AllLayers includes them.
• Extent::IsValid(): non-empty with all bounds finite.
• Test fixture MakeRectanglesDataset and MapTests.cpp (20 tests). No window and no ActiveX.
• docs/ARCHITECTURE.md gets a "Map model" section on identity, ordering, ownership, duplicates, visibility, extents, and coordinates.

Key design decisions

• Stable identity: LayerId is a monotonically increasing value that is never reused, not even after Clear. Positions shift; identities don't.
• Ordering: position 0 is the bottom layer and is drawn first. AddLayer appends on top.
• Ownership: MapLayer holds a shared_ptr<VectorDataset>, so the source layer outlives the caller's handle. Layers are stored as unique_ptr, so removing one layer never invalidates references to the others.
• Duplicates: the same source layer may be added more than once. Each addition gets its own identity and visibility.
• Extents are computed on each call rather than cached, so visibility changes and source edits never leave a stale extent. VectorLayer already caches its own extent.
• No reprojection: layers are assumed to share compatible coordinates, as documented.
• Pimpl classes keep the DLL ABI stable, in line with the existing Core types.

Dependencies / follow-up

• Based on development. Later phases stack on this branch.
• ActiveX is not switched to Map yet; that is Phase 3.

Build: Debug|x64, 0 warnings. Tests: 190 passed, 3 skipped (opt-in suites).

[Julian-MW]

Source review complete. I left two inline comments: one for the LayerId overflow edge case and one for missing coverage of a successful non-zero source-layer index.