merge inconsistent with merge.data.frame according to precision of numerical column

merge.data.table() and merge.data.frame() behave differently when the merge happens on numerical columns that differ on the last 1-2 decimal places:

zscore1 <- c(15.8068722584374992, -3.6299861879091400, 3.1547699616527298)
zscore2 <- c(15.8068722584374690, -3.6299861879091355, 3.1547699616527285)
dt1 <- data.table(id=10000012, allele="G", zscore=zscore1)
dt2 <- data.table(id=10000012, allele="G", zscore=zscore2, chrom=4)
merge(dt1, dt2)
# Empty data.table (0 rows and 4 cols): id,allele,zscore,chrom
merge.data.frame(dt1, dt2)
#         id allele    zscore chrom
# 1 10000012      G -3.629986     4
# 2 10000012      G 15.806872     4
# 3 10000012      G  3.154770     4
abs(zscore1 - zscore2)
# [1] 3.019807e-14 4.440892e-15 1.332268e-15

The level of precision used by either version of merge() is not documented, nor can be controlled. Such tiny differences can appear when interacting with a database (the first is written, and the second is read out), and possibly losing records on merge is unexpected.

# Output of sessionInfo()

R version 3.6.3 (2020-02-29)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Ubuntu 18.04.4 LTS

Matrix products: default
BLAS:   /usr/lib/x86_64-linux-gnu/blas/libblas.so.3.7.1
LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.7.1

locale:
 [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C
 [3] LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8
 [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8
 [7] LC_PAPER=en_US.UTF-8       LC_NAME=C
 [9] LC_ADDRESS=C               LC_TELEPHONE=C
[11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base

other attached packages:
[1] data.table_1.12.8

loaded via a namespace (and not attached):
[1] compiler_3.6.3

Start by running the supplied merge.data.table() and merge.data.frame() reproduction with the zscore values and comparing their matching behavior. Inspect the merge.data.table and merge.data.frame entry points for how numerical keys are compared, then determine and test the intended precision behavior; done should prevent records from being unexpectedly lost when values differ only in the shown tiny amounts.