fix performance regression caused by #32

the issue isn't with the pr itself, however it did expose the lack of caching with the translation strings, it will
continue to search for strings, even if we already found them.
to solve this, i used a map to cache said strings

sphaira/source/i18n.cpp

@@ -3,39 +3,54 @@
 #include "log.hpp"
 #include <yyjson.h>
 #include <vector>
+#include <unordered_map>
 
 namespace sphaira::i18n {
 namespace {
 
 std::vector<u8> g_i18n_data;
 yyjson_doc* json;
 yyjson_val* root;
+std::unordered_map<std::string, std::string> g_tr_cache;
 
 std::string get_internal(const char* str, size_t len) {
+    const std::string kkey = {str, len};
+
+    if (auto it = g_tr_cache.find(kkey); it != g_tr_cache.end()) {
+        return it->second;
+    }
+
+    // add default entry
+    g_tr_cache.emplace(kkey, kkey);
+
     if (!json || !root) {
         log_write("no json or root\n");
-        return str;
+        return kkey;
     }
 
     auto key = yyjson_obj_getn(root, str, len);
     if (!key) {
-        log_write("\tfailed to find key: [%.*s]\n", len, str);
-        return str;
+        log_write("\tfailed to find key: [%s]\n", kkey.c_str());
+        return kkey;
     }
 
     auto val = yyjson_get_str(key);
     auto val_len = yyjson_get_len(key);
     if (!val || !val_len) {
-        log_write("\tfailed to get value: [%.*s]\n", len, str);
-        return str;
+        log_write("\tfailed to get value: [%s]\n", kkey.c_str());
+        return kkey;
     }
 
-    return {val, val_len};
+    // update entry in cache
+    const std::string ret = {val, val_len};
+    g_tr_cache.insert_or_assign(kkey, ret);
+    return ret;
 }
 
 } // namespace
 
 bool init(long index) {
+    g_tr_cache.clear();
     R_TRY_RESULT(romfsInit(), false);
     ON_SCOPE_EXIT( romfsExit() );