Add `SPLIT_PENALTY_COMPREHENSION` flag.

Also: Remove the coarse flag on the comprehension state calculation and
make the flags determine whether/which penalties are added.

I've set the new flag's value to 2100 for Google as that seems to be a
pretty strong bias towards keeping everything on the same line (but not
too strong that it upsets common cases).

80 might be too high for PEP8 but I think 0 penalty results in some
pretty lousy formatting.
diff --git a/README.rst b/README.rst
index 8e155fd..96a648d 100644
--- a/README.rst
+++ b/README.rst
@@ -495,6 +495,9 @@
     The penalty of splitting the line around the ``&``, ``|``, and ``^``
     operators.
 
+``SPLIT_PENALTY_COMPREHENSION``
+    The penalty for splitting a list comprehension or generator expression.
+
 ``SPLIT_PENALTY_EXCESS_CHARACTER``
     The penalty for characters over the column limit.