fix: restore BigNumber precision state when decimal.js trig throws precision-limit errors

[mvanhorn]

Summary

Fix at the single choke point where mathjs creates its BigNumber class: src/type/bignumber/BigNumber.js (the factory that does Decimal.clone({precision: config.precision})). After cloning, wrap the trig-family prototype methods that use decimal.js's PI-based argument reduction (sin/cos/tan/asin/acos/atan and their cosecant/secant/cotangent/hyperbolic aliases as exposed by decimal.js: sine, cosine, tangent, etc.) in a thin guard that snapshots BigNumber.precision and BigNumber.rounding before delegating and restores them in a finally block, so a thrown DecimalError can no longer leave the class in a corrupted state. In the catch path, rethrow the DecimalError wrapped with an actionable message (e.g.

Why this matters

With math.config({number: 'BigNumber', precision: 509, ...}), evaluating tan(pi/2) (or any trig call on a high-precision argument) throws [DecimalError] Precision limit exceeded from decimal.js. The root cause is that decimal.js trig methods internally raise working precision to roughly precision + sd(x) guard digits for argument reduction against its PI constant, which is only stored to 1025 digits; 509 + ~509 exceeds that limit. Worse (per the reporter's follow-up comment), decimal.js mutates Ctor.precision/Ctor.rounding before computing and only restores them on success, so once the error is thrown the mathjs BigNumber class is left with a corrupted precision (~1025), and every subsequent BigNumber evaluation fails with the same error until the page is reloaded. The hard 1025-digit PI limit lives in decimal.js and cannot be lifted from mathjs, but the sticky global-state corruption and the cryptic error are mathjs-side bugs that can be fixed.

See #3676.

Testing

• Happy path: with default precision (64), tan(pi/2), sin(0.5), cos(bignumber(1)) still return correct BigNumber results and existing trig unit tests pass unchanged. - Sticky-state regression (core fix): configure a fresh math instance with {number: 'BigNumber', precision: 509}; assert evaluate('tan(pi/2)') throws, then assert a subsequent unrelated BigNumber evaluation (e.g. evaluate('1.1 + 2.2') or sin(bignumber(0.5))) succeeds and BigNumber.precision still equals 509. - Error path: the thrown error message mentions the precision limit of decimal.js trigonometric functions (not just the bare [DecimalError] Precision limit exceeded), and is an Error instance.

Fixes #3676