Another round of GCC fixes (#3115)

* Fix remaining <string> -> <cstring> for GCC compilation (#3114 follow-up)

MWerks' <string> header transitively includes C string functions
(memcpy, strlen, strcmp, etc.), but GCC/Clang's <string> is the C++
std::string header. These files all use C string functions and should
include <cstring> instead.

* Use std::isnan instead of isnan for GCC compilation

GCC's <cmath> places isnan in the std namespace. Using the unqualified
isnan fails to compile with GCC/Clang.

* Fix cCcD_Src types: s32 -> u32 for bitmask fields

cCcD_SrcObjCommonBase::mSPrm, cCcD_SrcObjTg::mType, and
cCcD_SrcObjAt::mType are used as bitmasks (SetType/SetSPrm take u32,
MskType/MskSPrm use u32, values like 0xFFFFFFFF are common in
aggregate inits). Change from s32 to u32 to match usage.

Also fix AT_TYPE_WOLF_ATTACK/AT_TYPE_UNK to use unsigned literals,
and remove now-unnecessary (s32) casts on hex literals in collision
source data.

* Mark dummy() functions as static to avoid multiple definition errors

These decomp artifact functions have the same name and signature across
TUs, causing linker errors when building as a single binary.

src/JSystem/JMessage/resource.cpp

@@ -4,7 +4,7 @@
 #include "JSystem/JGadget/search.h"
 #include "JSystem/JGadget/define.h"
 #include <algorithm>
-#include <string>
+#include <cstring>
 
 u16 JMessage::TResource::toMessageIndex_messageID(u32 uMsgID, u32 upperHalf, bool* pbValid) const {
     if (!oParse_TBlock_messageID_.getRaw()) {