Tree initialisation operators sometimes don't achieve stated depth

[jmmcd]

While investigating #140, I have seen both rhh and PI_grow returning individuals with depth less than min depth. For example I'm using:

<solution> ::= <rule>
<rule> ::= <rule> and <bool-expr>|<bool-expr>
<bool-expr> ::= p<idx> > p<idx>|p<idx> < p<idx>
<idx> ::= 0|1|2|3|4|5|6|7|8|9|10|11|12

and python ponyge.py --population_size 100 --grammar ../grammars/test.bnf --min_init_tree_depth 8 --max_init_tree_depth 10.

When generating with depth 9 I get some individuals of depth 7, when generating depth 10 I get some individuals of depth 8.

I guess in other cases there could be grammar constraints that eg allow depth 8 but prevent depth 7, or vice versa, but I think with this grammar there aren't.