Fixes windows error on large MAZ systems

[jpn--]

Turns out the error in #681 has actually been a longstanding problem, since the dawn of the use of MAZ's in ActivitySim. It's just that it has only come to light recently as it requires a particular confluence of conditions to have it cause the crash that PSRC was seeing:

1. running on Windows (hence default integer type is int32, not int64)
2. using a two-zone system with more than 32K microzones (hence some combination of origin and destination zone index codes will overflow an int32)
3. a specification that prohibits trip modes other than WALK when the tour mode is WALK
4. a tour generated with a tour mode of WALK, which goes to a destination that is accessible by WALK according to the MAZ-level skims, but would not be accessible by WALK if only the TAZ level skims were used
5. Either the origin or the destination zone index code of this tour is larger than 32768.
6. Also, whatever zone pair happens to match the overflowed remainder must not have a valid WALK travel time defined.

Obviously this is a very narrow corner case to cause a crash. However, the bug was (slightly) more serious than that. All these conditions need to be met to cause a crash, but only a subset are needed to cause erroneous results: MAZ skim values on Windows were being incorrectly evaluated for many trips and tours where the origin or destination zone index code is larger than 32768.

This fix upcasts the index lookup values to 64 bit when necessary, solving the problem.

Many thanks to @stefancoe for his assistance in highlighting and correcting this insidious error.