This re-allows parsing partial descriptors, which had previously been allowed until
7672fb5221.

a0ceb18c89 broke this.

For some case the encode function was raising AssertionError despite
the BCD value beeing correct (in this case it was 1.11).

There is two main modification in the version I propose:

First the added rounding before converting the "percent" to an integer.
This may seems odd but it is actually to "correct" the error introduce
by the multiplication.

Second the comparison is done "after" conversion by recreating the
floating point value from the integer and percent variables. This is to
make sure that the value in the descriptor is the same as intended.

I've made sure that all correct values (that can be stored in bcd fields)
doesn't raise an error when passed to the encode function.

I've used the following program to test the new implementation:
```
for obj in [float(f"{i:02}.{p:02}") for i in range(100) for p in range(100)]:
    integer = int(obj) % 100
    percent = int(round(obj * 100)) % 100
    if float(f"{integer:02}.{percent:02}") != obj:
        print (f"BCD conversion error: {integer:02}.{percent:02} != {obj}")
```