Constructor loses precision when parsing small numbers expressed in scientific notation ("e" format).

[StrangeRay]

The problem is perhaps best explained by example. Consider the construction of the decimal 1.23456789×10⁻²¹ via decimal("1.23456789e-21"). The result should display as
Decimal(0, 123456789, -29) but I get Decimal(0, 0, -20) instead. Observe the progressive loss of significance in the following constructions:

julia> using Decimals

julia> decimal("1.23456789") # Ok
Decimal(0, 123456789, -8)

julia> decimal("1.23456789e-11") # Ok
Decimal(0, 123456789, -19)

julia> decimal("1.23456789e-12") # Still Ok
Decimal(0, 123456789, -20)

julia> decimal("1.23456789e-13") # Not quite
Decimal(0, 12345679, -20)

julia> decimal("1.23456789e-14") # Constructor loses one significant digit per decade
Decimal(0, 1234568, -20)

julia> decimal("1.23456789e-15") # Trend continues ...
Decimal(0, 123457, -20)

julia> # ...
       
       decimal("1.23456789e-18") # ...
Decimal(0, 123, -20)

julia> decimal("1.23456789e-19") # ...
Decimal(0, 12, -20)

julia> decimal("1.23456789e-20") # Down to one significant digit
Decimal(0, 1, -20)

julia> decimal("1.23456789e-21") # All significant digits are lost
Decimal(0, 0, -20)

The problem is traceable to the implementation of function scinote. In particular, see line 39 in file decimal.jl.