Desired behavior

Ensure all objects emit floating point numbers of same precision.

At present, there may be different (default?) precisions in use by composite types in ign::math, as encountered by:
#689 (comment)

Alternatives considered

N/A

Implementation suggestion

Consider setting precision in ostream object inside sdformat, which is then (hopefully) used directly by composite types.

Additional context

N/A

\cc @azeey