Conversion from double to float when counting differences

[matthew-white]

For float variables, do not count a value as a difference even if it differs from the value in the replacements file as long as the two values' float representation is the same. Example replacements file for the auto dataset:

make	varname	value
Volvo 260	gear_ratio	2.9800001

E.g., the following code creates this replacements file, then feeds it to readreplace:

clear
set obs 1
gen make = "Volvo 260"
gen varname = "gear_ratio"
gen double value = 2.9800001
tempfile rf
sa `rf'
sysuse auto, clear
tempfile auto
sa `auto'
assert gear_ratio == float(2.9800001) if make == "Volvo 260"
readreplace using `rf', id(make) var(varname) val(value) use
cf _all using `auto'

There is no actual replacement to gear_ratio, which we confirm with cf, because even though gear_ratio != 2.9800001 in that observation, gear_ratio == float(2.9800001). Yet readreplace displays that there is 1 replacement. This is wrong and should be fixed.