parse derivatives instead of just relying on tree

[drdrew42]

Taking derivatives using the D method ends up creating formulas that refuse to reduce (e.g. constant products appearing in the derivative of a polynomial).

This PR modifies the behavior of the D method -- preserving the existing tree structure of the derivative, but now processing the result of that process (respecting the existing context flags for free).

[dpvc]

This line means that any constants are clipped to 6 significant digits (the default number format), so the result may have different from the original.

You may also want to add a new reduction rule that controls doing this doing this, so it can be turned off, if desired.

[Alex-Jordan]

I think we can close this. The specific instance cited (e.g. constant products appearing in the derivative of a polynomial) has already been addressed by #1025. Is there an example of something different that this would still help with?

In case this should be kept, replying to @dpvc's comment: could the string method at the new line 43 take an argument that briefly changes the number format to use more precision? Say 17 significant digits?

[dpvc]

The string methods already take optional parameters, so you can't use that approach to adjust the precision. But there is a context flag for the number format, and an expression can override that using its format property, so you could use

	my $deriv = $self->new($f);
	$deriv->{format} = '%.17g';
	$deriv->{string} = $deriv->{tree}->string;
	delete $deriv->{format};

in order to get the higher precision. I doubt it is necessary, but you could save any old format value and replace it, if there is one.