Description:

fn day_timestamp_micro (source) seems to produce the wrong result which is off-by-one due to rounding errors.
Also the implementation does not align with the Java implementation, which handles the conversion slightly different.

An example of the wrong output can be seen in the test:
-115200000000 evaluates to 1969-12-30T16:00:00Z which should be days -2 (since epoch) and not -1.

Proposal:

Align Rust with Java implementation.
I already have a prototype, due to working on #264, so I will implement the fix in a different PR - so it can be verified more easily.