demux: mp4: enforce string formatting consistency in MP4_BoxGet/MP4_BoxCount

Moving the string formatting in the .h helped find the issues fixed here. However it forces to use %s where the call doesn't use a string literal.

---

v2: only use MP4_BoxGetVa() in the rare cases a vararg is needed. Otherwise we avoid a string parsing+allocation. MP4_BoxCountVa() is added but never used, in case it's needed someday.