fix(csvbase): do not persist auto-detected self.order

[wzyboy]

Today when I batch import two years of statements for a rarely-used account, I find this auto detection of row order might produce incorrect results.

For example, there is only one transaction in January:

2023-01-31,...

... and four transactions in February:

2023-02-28,...
2023-02-20,...
2023-02-11,...
2023-02-05,...

When processing the January file, the code auto-detects the order as ASCENDING as there is only one transaction, and for Feburary and any subsequent files, the code no longer tries to detect the order because self.order is already set. But this bank actually uses DESCENDING order for their files.

This error might be masked if the entries are eventually sorted. But I disabled sorting for some other reason so I hit this edge case.