Store tags in one big pre-allocated string and only work with refs

This should decrease the number of allocations before main is entered
significantly, but complicates the code somewhat in return.

Assuming I used `massif` right, doing just `SelfTest --list-tests`
went from 929 allocations at "Remove gcc-4.9 from the travis builds"
(2 commits up), to 614 allocations with this commit.

include/internal/catch_test_case_registry_impl.cpp

@@ -78,6 +78,10 @@ namespace Catch {
         m_invokers.push_back(std::move(testInvoker));
     }
 
+    std::vector<std::unique_ptr<TestCaseInfo>> const& TestRegistry::getAllInfos() const {
+        return m_infos;
+    }
+
     std::vector<TestCaseHandle> const& TestRegistry::getAllTests() const {
         return m_handles;
     }