Added fixes for QIF files with different date order and syntax

[jmcameron]

I ran into difficulties with QIF files exported by an old version of Quicken. Some of the numbers had embeded commas (for thousands separators) and the date as in DD/MM'YY format instead of MM/DD'YY format the code expects. I added a couple of class-level flags for this. Normal use should not change, but By giving extra (optional) arguments to the .parse() command, it handles my odd formats too.

I changed the Decimal(str) use to use a special function that removes the embedded thousands separators (commas by default, but overrideable) before doing the conversion to a flow. I only did it where I had difficulties (processing transactions); you may want to change the rest correspondingly.

[jmcameron]

Converted more Decimal() calls to convertFloat() calls for values that should be floats.
Fixed issue to get test_parser regtests working (by converting transaction amounts back to strings with xxx.nn format. My guess is this has been broken since adding float() conversions (and stayed broken with Decimal() changes). The regtests pass for me. It would not hurt to add more tests with different flavors of QIF files.

[jmcameron]

Regtests now pass on Travis CI!

[jmcameron]

In the last update to this pull request, I added support the 'U' field (seems like another amount for transactions) and named it 'amount_U' since I could not find any documentation on it.