Time shift (many months) in time conversion after importing data (csv-import, CLI, line protocol ...)

__Steps to reproduce:__
This seems to be a principle problem, but one scenario is very easy to reproduce

1.) Go in the web-frontend to Data->Line Protocol to enter data manually

2.) Use exactly the example string mentioned below

1556813561098000000 should represent **Thu May 02 2019 16:12:41 GMT+0000**
3.) Go to the data base an verify the time

You can see that the time is not correct **(2019-07-18T05:04:10.000Z)**

4.) Make an csv export of this data


5.) Import the same file to the same bucket

6.) You can see that there has been another time inserted

**Thu May 02 2019 16:12:41 GMT+0000 --> 2019-07-18T05:04:10.000Z --> 2019-11-20T00:08:20.000Z**

__Expected behavior:__
Importing time values from the past should work (without a time shift)

__Actual behavior:__
There is a problem converting the time.
Import data without time works fine (uses actual time)
The problem must be general because I detected it with csv import of an excel sheet via CLI.

__Environment info:__
*Running on Windows 10
*InfluxDB 2.1.1 (actual released version)

Reproduce the timestamp conversion through the web frontend's Data→Line Protocol flow, then compare the CSV export/import and CLI or line-protocol ingestion paths. Trace where the supplied timestamp is parsed and converted; done means the example remains Thu May 02 2019 16:12:41 GMT+0000 through each path without shifting, including re-import.

Written by the indexing model from the issue text.