Date.prototype.setTime must store the clipped time value

[lahma]

Date.prototype.setTime stored the unclipped argument:

var v = t.TimeClip();
((JsDate) thisObject)._dateValue = t;   // spec stores v
return v.ToJsValue();

21.4.4.27 step 5 sets [[DateValue]] to v. All 15 sibling setters already do this correctly — setTime was the sole outlier.

The effect is larger than "an infinity survives". DatePresentation's conversion turns ±Infinity into a flagged presentation with Value = 0, and ToJsValue() tested only IsNaN, so an infinity read back as the epoch:

var d = new Date(); d.setTime(Infinity); d.getTime();   // 0, should be NaN
d.setTime(8.64e15 + 1); d.getTime();                    // NaN, but...
d.toISOString();                                        // "+275760-09-13T00:00:00.001Z"
d.setTime(Number.MAX_VALUE); d.toISOString();           // "+292278994-08-17T07:12:56.807Z"

The last two are finite-but-out-of-range, so they are not fixed by a toISOString guard alone — only storing v fixes them. (Number.MAX_VALUE produced year 292278994 because the double->long cast saturates.)

The fix

Three lines:

1. setTime stores v.
2. toISOString guards on "not finite" rather than "is NaN".
3. DatePresentation.ToJsValue() answers NaN for any non-finite presentation, not just a NaN one.

On (2): 21.4.4.36 step 4 literally says "If tv is NaN, throw a RangeError", and in the spec that is the whole of "not a finite time value" — a [[DateValue]] is by construction either NaN or a finite in-range integral Number. DatePresentation is wider than that domain, so IsFinite is the same step expressed over the representation Jint actually has. The code comment says exactly this rather than misquoting the spec.

(3) closes the class rather than the instance. Every ToJsValue consumer was checked; the only behavioural delta is IsInfinity reading back as NaN instead of 0. In particular the DateTimeMinValue/DateTimeMaxValue flags are not in the finite mask but their values sit well inside the TimeClip range, so they still read back as numbers — pinned by a test.

Tests

Jint.Tests/Runtime/DateTests.cs — 30 cases, 9 failing against unfixed code, covering setTime with ±Infinity/NaN/8.64e15 + 1/Number.MAX_VALUE, the exact 8.64e15 boundary, toISOString raising RangeError, and setTime agreeing with the subsequent getTime/valueOf. No existing expectation changed.

Frees two staging/ exclusions: Date/timeclip.js and Date/toISOString.js.

Refs #3021.