Looking into #50237 I realized that two of the examples given in the
documentation around date math rounding for range queries on date fields using
`gt` and `lt` is slightly off by a nanosecond. This PR changes this to the
bounds that are currently parsed using these parameters.