Construction of canonical exponential strings is overly cumbersome

https://github.com/tc39/proposal-amount/pull/112#discussion_r3270010468

I see four uses of RenderInExponentialNotation:

• in the Amount constructor, with the fields of an Intl MV Record derived from a FormatNumericToString output string
• in Amount.prototype.convertTo (here), with the fields of an Intl MV Record derived from a FormatNumericToString output string
• in Amount.prototype.toString, with the fields of an Intl MV Record derived from a Number
• in Amount.prototype.toString, with the fields of an Intl MV Record derived from a BigInt::toString(v, 10) decimal digit sequence (where those fields could alternatively be defined directly by counting the digits in the decimal representation of _v_, e.g. ceiling(log10(abs(_v_) + 1)) or a named equivalent)

The uses of FormatNumericToString seem to be necessary, but it occurs to me that we could probably remove a lot of translation, e.g. carrying non-exponential decimal strings from FormatNumericToString and maybe-exponential decimal strings from Number::toString all the way to the exponent-ensuring and decimal-point-shifting work of RenderInExponentialNotation without passing through non-string intermediaries.

Start with the linked PR discussion and trace the four uses of RenderInExponentialNotation in the Amount constructor, Amount.prototype.convertTo, and Amount.prototype.toString. Compare the FormatNumericToString, Number::toString, and BigInt::toString paths; done means reducing unnecessary translation while preserving canonical exponential-string behavior across all listed call sites.