RDataFrame tree conversion takes too much time and memory

Hi Enrico,

I’ve tested my code on the same Ubuntu machine with ROOT’s master version. The run times in seconds are

python, "heavy"	python, "light"	cpp, "heavy"	cpp, "light"
12,6357138157	21,2765231133	10,8265318871	20,9584360123
8,6492950916	16,8159852028	8,8113510609	17,554530859
8,3732161522	17,013764143	8,5724999905	18,1778972149
8,4610950947	18,0920848846	8,7197887898	18,6586410999
8,4218490124	17,5790328979	8,8868849278	18,871502161
8,6015489101	16,7283360958	8,7004339695	19,2456002235
8,7279579639	17,1717660427	9,1521298885	18,8681650162
8,5640778542	17,6466178894	8,8281259537	18,9460320473
8,6654288769	17,0789301395	9,1617879868	18,6554582119
8,7749049664	17,5882809162	8,6360168457	18,7109069824
8,5787391663	18,0985980034	8,6626999378	18,7014679909
8,8500330448	20,2722699642	8,7691810131	19,0996360779
8,6696448326	22,8912670612	9,2500059605	19,285490036
8,7564928532	20,6616809368	9,3329889774	20,0462639332
8,7350840569	18,5314588547	9,0782361031	19,4825839996
8,8899960518	18,2859280109	8,9463450909	19,6934659481
8,7146229744	18,0480101109	8,9019899368	18,2985579967
8,7302470207	18,2739129066	9,2911269665	17,2530460358
8,9851078987	20,4818389416	9,6494119167	19,7338588238
9,1543111801	20,0699179173	9,5622749329	20,7159221172

It’s slightly faster compared to v. 6.16/00 but still not as fast as pure C++ tree conversion code (like I’ve checked here.
Memory issues are still present too.