DecimalNum: reduce DEFAULT_PRECISION from 32 to 16 or less

[nimo23]

Despite the discussion about #916, what we really can do to increase the default performance immediately and noticeably, is to decrease the precision of DecimalNum#DEFAULT_PRECISION:

public final class DecimalNum implements Num {

    private static final int DEFAULT_PRECISION = 32;

to

public final class DecimalNum implements Num {

    private static final int DEFAULT_PRECISION = 16; // or less?

Reasons:

• Too high a precision leads to excessive computational effort and has de facto no added value. If so, please prove it!
• Most cryptocurrency trading platforms and analysis tools typically operate with limited accuracy, often to 8 or 16 decimal places.

[nimo23]

Looking for example, at the smallest unit of ETH (=wei) which is equivalent to 10^-18 ETH ( = 0.000000000000000001 ETH). So maybe using a precision of 20 instead of 32? I don't know if this matters to choose a far smaller one. For example, in DoubleNum we only use a default precision of 6 digits.

It would be also good to use the same default precision for both DecimalNum and DoubleNum, i.e. use a precision of 12 (or 20?) for both Num-types - with this, we can even go further were it really does matter to use DecimalNum, where e.g. "DecimalNum" should really be preferred or vice versa. It didn't make sense to have different precisions for both num types in the first place, since there could never be an equal choice between them. To sum up: It makes sense to use the same precision for both kinds of Num-ypes, otherwise the user doesn't have an equal choice between them.

[TheCookieLab]

I'm open to exploring this further, although am confused by the line:

Too high a precision leads to excessive computational effort and has de facto no added value. If so, please prove it!

Is this a prompt to yourself to "prove" the value of this change?