fix: use total_seconds instead of seconds in the datetime

[kan-fu]

Using seconds is clearly a bug. Running

import datetime
print(datetime.timedelta(days=1, seconds=13599).seconds)
print(datetime.timedelta(days=1, seconds=13599).total_seconds())

will print 13599 99999.0. We would like 99999 in our case.

Also add more message in the log. Now it looks like

Data quantity is greater than the row limit and will be downloaded in multiple pages.
Downloading time for the first page: 8 seconds
Estimated approx. 6 pages in total.
Estimated approx. 40 seconds to complete for the rest of the pages.

(50000 samples) Downloading page 2...
(100000 samples) Downloading page 3...
(150000 samples) Downloading page 4...
(200000 samples) Downloading page 5...
(250000 samples) Downloading page 6...
(259201 samples) Completed in 40 seconds.

The calculation is based on the response time of downloading the first page. This log makes it clear that the estimated 40 seconds is calculated by 8 seconds/page * (6 - 1) page.
The estimated pages will still be inaccurate if the specified time span includes date range that has no data. This is unavoidable because it cannot be known in advance until the request is made. I found that when running the code below in QA. There are no data after 2019-11-27 for this device in QA.

from onc import ONC
onc = ONC("XXX", production=False)
onc.getDirectByDevice({
        "deviceCode": "BPR-Folger-59",
        "dateFrom": "2019-11-26T00:00:00.000Z",
        "dateTo": "2019-11-30T00:00:00.000Z",
        "rowLimit": 50000,
    },allPages=True)

Data quantity is greater than the row limit and will be downloaded in multiple pages.
Downloading time for the first page: 8 seconds
Estimated approx. 7 pages in total.
Estimated approx. 48 seconds to complete for the rest of the pages.

(50000 samples) Downloading page 2...
(86400 samples) Completed in 15 seconds.

Test failures in the Actions is caused by the multiple page bug in the server. It will be fixed in the next minor release from the server side.