Fix a bug that an SGF file/string cannot contain 2 consecutive moves of the same color

[zhanzhenzhen]

Fixes #1469.

As I'm not a C++ developer, please check my modifications carefully to see if they are all correct. I've tested, but maybe there're coding style or other problems.

This bug will be triggered if:

• There're 2 or more consecutive moves of the same color; or
• There's AW and the first move is WHITE; or
• There's no AW but there's AB, and the first move is BLACK.

The reason is that SGFTree passes get_to_move() as the argument when calling get_move method, so if all moves are not "black, white, black, white...", then it won't be loaded correctly. (Here all handicap/setup stones can be considered as one move and the color can be considered as black if there's no AW, or white if there's AW.)

I've replaced 2 places of get_move. There's still one place left, but that's related to training so it's impossible to trigger this bug (but maybe better also change it?).

The follow_mainline_state modification is required. But during my test, I found populate_states modification takes no effect, that is, it seems also OK if unchanged.

Another thing I don't know why is that std::make_pair(FastBoard::INVAL, 0) works, but if modified to std::make_pair(FastBoard::INVAL, SGFTree::EOT) then it will report error when compiling.

[gcp]

Another thing I don't know why is that std::make_pair(FastBoard::INVAL, 0) works, but if modified to std::make_pair(FastBoard::INVAL, SGFTree::EOT) then it will report error when compiling.

Seems to work for me. I did some minor cleanups here: #1654.

[gcp]

Actually, the compiler error was likely caused because this was missing:
7524a2f#diff-44ba5da097788068a1c8def978c3293cR40