[BREAKING] chore: adopt proposed new config option organization

[gwhitney]

This PR:

• deprecates most existing config options, allowing for a translation table into the newly-supported options
• reorganizes almost all options into one of three groups: parse, compute, or compatibility, to indicate what aspect of the system it controls. The only exception is the number option giving the global default number type, as it affects both parsing (math.evaluate('2 + 2')) and computation (math.sum([])), per the proposal in issue 3420. So that option is not deprecated and remains at the top level.
• adds the compute.numberApproximate option to control the type of output when the config.number option is not suitable. Currently this is only used for physical constants (since there is no input on which to base their type, and there is no guarantee that config.number will support their values), but it is planned for use with square roots of bigints, for example.
• validates all config options and adds warning for unrecognized options
• updates all documentation and tests and typescript typing
• conforms the behavior of statistics functions min, max, sum, and prod with all other mathjs functions at the moment so that strings are interpreted by way of typed-function's conversion rules, rather that mathjs' parsing rules. (There is a proposal to switch to the latter across the board; but in the meantime, mathjs functions should be consistent with each other.)
• adds the zero() function (to streamline the update of isPositive and isNegative; it was pending on another PR anyway)
• implements unaryMinus correctly on boolean, to match the implementation of unaryPlus
• Prevents simplifyConstant from changing intentional Fraction constants back into the default number type, potentially introducing roundoff errors. For this sake, it internally "advises" some of the Fraction methods so it can track whether a given Fraction in the result came from Fractions that were autogenerated from numbers (and so should be converted back to the default number type) or were in the original expression as Fraction constants. All this advice is removed before simplifyConstant returns. If the reviewer is uncomfortable with this approach, there are alternatives that do not touch the Fraction prototype even temporarily. For example, we could create an explicit datatype used for the folded constants that externally, rather than internally (by a property on the Fraction object) tracks whether they came from non-Fraction number constants. It would just be significantly more new code in simplifyConstant.js. You might reasonably ask how this change became entangled with this particular PR -- it's because the number-type config parameters have changed, and the old ones were used in selecting how to turn a Fraction into other types at the end of simplification, and in trying to update that code, I discovered the bug that actual Fractions in the original expression could be transformed into (say) JavaScript number in the result. It seemed like an opportune time to fix the bug.
• adjusts the docgenerator so that the list of config options, which is already long and will be getting longer, only has to be documented in one place (the docs/core/configuration.md list is copied into the reference/function/config.md page). This will ease adding new options and prevent the inevitable drift of differences (that was already underway) between the two locations.
• Adjusts the create() function so that it initially creates with the default config and then loads the supplied config, so that the supplied config can undergo the same option translation and validation processes as happens with explicit calls to the config() function.
• discontinues the previously-deprecated epsilon option; note the option is still noticed and throws a discontinuation error if someone tries to set it.
• deprecates the config.compatibility.subset option so that in 2+ future major versions it can be discontinued.
• eliminates the config.MATRIX_OPTIONS and config.NUMBER_OPTIONS properties in favor of a comprehensive config.ALLOWED_CONFIG property that shows all allowed options and the validity condition on each one.
• removes the config dependency from several functions that were not actually using it
• Downgrades to Complex number precision on out-of-real-range trigonometric functions on BigNumber arguments, rather than just returning BigNumber(NaN), to conform with all other functions (e.g. log).

As reminders, the rationales for this option reorganization are:

• To pave the way for a host of parsing configuration options to move off of the parse() function and into the config.parse option group, so that they will be more clearly documented and exposed to the library users, and cannot be changed on the library static instance (cf. #3420)
• To make room for per-type tolerance parameters with systematic notation in the config.compute.TYPENAME groups, including for the built-in number type, without clashing with the top-level number option (cf. #3556)
• To provide for a minimal collection of type-choosing parameters that cover known use cases: config.number at the top level affecting both parsing and computation, config.parse.numberFallback for interpreting decimal literals when the config.number type does not support them, and config.compute.numberApproximate for the return types of functions when the input does not necessarily determine the type and the result may not be representable as config.number (e.g., math.evaluate(sqrt(5)) when config.number() is bigint).

When this is reviewed, adjusted as necessary, and merged into v16, work can resume on (a replacement for) #3423, which will (for example) systematize the tokenization of operators like ?. that end in a dot, as well as allow new token types (such as #HHHHHH for color constants) on the part of library users.