Fix various -Wmaybe-uninitialized warnings from GCC 12.2.1

[akien-mga]

Not sure why I didn't get those before, it may be due to upstream changes (12.2.1 is a moving target, it's basically 12.3-dev), or simply rebuilding Godot from scratch with different options.

Edit: I got those while testing #66242, I think the difference is that dev builds now default to -Og optimization on Linux (debugging) instead of -O0 (no optimization). GCC finds more/different warnings with optimizations enabled.

[akien-mga]

Please review @godotengine/rendering @TokageItLab.

I don't know if that's the right fix but reading the code it does look like we're potentially dereferencing an uninitialized pointer here since it's only initialized for RS::ARRAY_VERTEX, RS::ARRAY_COLORS and RS::ARRAY_BONES but this loop goes through all of RS::ARRAY_MAX.

[akien-mga]

Added a couple warning fixes to be able to build with tools=no target=release. I've been seeing those on and off for a while but always started overengineering the fix, now I've just done the minimum.

There's also a series of warning in internal BVH code that need to be fixed to build release templates without warnings, @lawnjelly is looking into it. This was my workaround:

diff --git a/core/math/bvh_structs.inc b/core/math/bvh_structs.inc
index 58c8f0479a..82e5bde870 100644
--- a/core/math/bvh_structs.inc
+++ b/core/math/bvh_structs.inc
@@ -61,20 +61,20 @@ private:
 public:
 	// accessors
 	BVHABB_CLASS &get_aabb(uint32_t p_id) {
-		BVH_ASSERT(p_id < MAX_ITEMS);
+		ERR_FAIL_UNSIGNED_INDEX_V(p_id, MAX_ITEMS, aabbs[0]);
 		return aabbs[p_id];
 	}
 	const BVHABB_CLASS &get_aabb(uint32_t p_id) const {
-		BVH_ASSERT(p_id < MAX_ITEMS);
+		ERR_FAIL_UNSIGNED_INDEX_V(p_id, MAX_ITEMS, aabbs[0]);
 		return aabbs[p_id];
 	}
 
 	uint32_t &get_item_ref_id(uint32_t p_id) {
-		BVH_ASSERT(p_id < MAX_ITEMS);
+		ERR_FAIL_UNSIGNED_INDEX_V(p_id, MAX_ITEMS, item_ref_ids[0]);
 		return item_ref_ids[p_id];
 	}
 	const uint32_t &get_item_ref_id(uint32_t p_id) const {
-		BVH_ASSERT(p_id < MAX_ITEMS);
+		ERR_FAIL_UNSIGNED_INDEX_V(p_id, MAX_ITEMS, item_ref_ids[0]);
 		return item_ref_ids[p_id];
 	}
 

[akien-mga]

The alternative is to initialize this variable before the switch even though it should be properly initialized in all switch branches.

The problem as I understand it is that enums suck :) So even if we make sure only those three values are passed as input, in theory something could call with a garbage int and it would work just fine, using uninitialized texture.rd_type.