fix(date): ECMAScript ToNumber for Date arguments (Symbol/BigInt throw, single valueOf)

[proggeramlug]

Summary

Date argument coercion bypassed ECMAScript ToNumber: for any non-string / non-sentinel value the helpers read the raw NaN-boxed bits as a double, so a Symbol/object/array/boolean produced garbage or an Invalid Date instead of throwing or coercing.

Before → after (vs node --experimental-strip-types)

new Date(Symbol())        Invalid Date   →  TypeError          ✓
d.setHours(Symbol())      (no throw)      →  TypeError          ✓
Date.UTC(2020, Symbol())  (no throw)      →  TypeError          ✓
new Date(true)            Invalid Date   →  1                  ✓
new Date(2020, {valueOf}) valueOf 0×     →  valueOf exactly 1× ✓

Fix

Routes the shared jsvalue_to_number helper, the new Date(v) numeric branch, and the multi-arg new Date(y,m,d,…) coerce through one ECMAScript ToNumber:

• Symbol / BigInt → TypeError (not numerically convertible);
• Date → its time value;
• object / array → a single valueOf/toString via OrdinaryToPrimitive(number) (so valueOf is invoked exactly once), then the primitive numeric coercion;
• string / boolean / null / undefined → the existing ordinary coercion.

All in date.rs (no new file; under the line cap).

Results

test262 built-ins/Date: 202 → 158 (44 fixed, no regressions), measured uncapped via scripts/parity_gap_report.py. Verified Node-identical in both PERRY_NO_AUTO_OPTIMIZE=1 and default builds. cargo test -p perry-runtime --lib: 979 passed, 0 failed.

The remaining Date failures are separate root causes (a follow-up will add Date.prototype method this brand-checks, ~52 cases; plus arg-vs-date-value read ordering and toISOString RangeError edges).