refactor(spx): centralize coordinate conversion

[joeykchen]

goplus/spx#1702
goplus/spx#1703

[fennoai]

Review: refactor(spx): centralize coordinate conversion

Solid, well-scoped refactor. Centralizing the Y-axis flip into spx_to_godot_vec2 / godot_to_spx_vec2 and applying it at the manager API boundary removes a lot of error-prone scattered y = -y / flip_y() code, and the get/set pairs are now symmetric. The RenderRoot restructuring and the two new unit-test headers are good additions. The security pass found no memory-safety issues (null-guards, batch-buffer bounds checks, and reparent lifecycle all hold up).

One finding is worth confirming before merge because it is a semantic change hiding inside a "refactor", plus a few smaller items — see inline comments.

Summary of findings

• Collider/trigger center & polygon-vertex Y-flip changed (see inline on spx_sprite_collision.cpp). This is the main item: several shape setters now flip the center Y (and collider-polygon vertices) where the base code did not. The old code was internally inconsistent, so the new unified behavior may well be the intended fix — but it is not a no-op and can vertically mirror collider/trigger offsets in existing SPX content. Please confirm it is intentional, document it, and ideally add a test asserting center placement per shape type.
• set_render_offset re-positions RenderRoot on every batch update (inline on spx_sprite_render.cpp) — a change-guard would avoid redundant transform-dirty propagation (and the dynamic-frame-offset recompute) on the per-sprite batch hot path.
• godot_to_spx_vec2 is an undocumented alias of spx_to_godot_vec2 (inline on spx_coordinate.h).
• get_rect() null-render_root fallback silently returns a differently-origined rect (inline on spx_sprite_texture.cpp).

Additional (no inline location):

• Naming inconsistency: SpxSprite renamed pivot_offset/set_pivot → render_offset/set_render_offset, but the sibling classes SpxRenderSprite/SpxStaticSprite (and spx_scene_mgr.cpp:225, the gdextension_spx_ext bindings) still use pivot/pivot_offset for the same concept. Functionally fine (separate code path), but a follow-up to unify the vocabulary — or a short comment noting the intentional divergence — would help readability.