omegaone, it looks like I have the same shunt as you, so I have tried it in my Fluke 8846A, the meter was factory calibrated in April 2018.
In 2 wire mode, the reading was 0.00549 Ohms. In 4 wire mode, this dropped down to 0.00389 Ohms. The physical setup was the same, I just switched between the 2 wire and 4 wire mode. Having the short connected directly as per photo 3, the reading was 0.00489 Ohms. I did not get a negative reading at any point, but the ohmic values seem to be similar to yours.
Playing around with the instrument, I did get a negative reading in 2 wire, if I compensated and in 4 wire if I reversed the sense connections.
I then tested a 1mOhm shunt 0.1% accuracy. Again no negative ohms reading was obtained, although the readings were no where near where they should have been. The 2 wire measurement was high, which is what I would have expected, the 4 wire was low, which is not expected, so there may have been some setup error causing this.
A 10mOhm shunt produced a high reading in 2 wire mode, and a much closer reading in 4 wire mode. Which is pretty much what I would have expected, although given the tolerance of the 10mOhm shunt, also 0.1%, the 4 wire reading is too high.
Kind regards.