A different thought:
I used the LINEST formula on your data to get the slope and intercept and it appears that it treats the Y axis as the date 1/1/2001. Every other date is "days since, or before, 1/1/2001". The trend line equation on the chart is somewhat different. In it, the other dates are "seconds since 1/1/2001 12:00AM".
The problem you were having with the actual dates and the problem now with your numeric "dates" is due to rounding of the slope and intercept in the formula on the chart. The numbers being used for X are very large compared to the minuscule percentage values for Y. You will have to calculate the slope and intercept yourself in your table using LINEST to get the precision you need.
And for some additional info that I already typed up before the stuff I added above:
One date structure of Excel uses 12/31/1899 as day 0. I often try to stay consistent with Excel but this time it is a problem. Numbers counts 1900 as a leap year when it is not. So if you use 12/31/1899, your index numbers will not match Excel's. I wonder if Numbers gets other non-leap years wrong. 1900 is a century year not divisible by 400 so is not a leap year. I have not tested any others. To be consistent with Excel for dates after 2/28/1900, you would need to use 12/30/1899 as the starting date.
I had no issues using "1/1/0001" in DATEDIF and I am getting different results than you.
=DATEDIF("1/1/0001",C2,"D") where C2 has 3/15/2021 was not an error. It gave me the value 737865
737865+366 = 738231 (not the 738229 that you got)
