Add interrupt checks to more Array methods

[yspbwx2010]

Follow-up to #1674 (issue #1672), which added js_poll_interrupts() to the scan loops of the iterating Array methods. A number of other Array loops walk [0, length) in C without calling into JS, so they still never consult the interrupt handler. On a sparse array or an array-like with a large length, Ctrl-C in the REPL has no effect on them. For example, Array.prototype.join.call({length: 2**32-1}, "") runs for four to five minutes, and Array.prototype.slice.call({length: 2**32-1}) and new Array(2**32-1).reverse() behave the same way.

This adds a poll at the top of those loops:

• js_array_join (join, toLocaleString)
• js_array_reverse, generic path
• JS_CopySubArray (copyWithin, shift, unshift, splice)
• js_array_slice, the generic copy loop (slice, splice) and splice's delete loop
• js_array_concat, the loop over a spreadable argument
• js_array_sort, the gather loop and the loop that deletes trailing holes
• JS_FlattenIntoArray (flat, flatMap)
• js_array_fill
• js_array_from, the array-like path

The check is the same as in #1674. Each error exit follows the function's existing one:

• goto fail in js_array_join, so the string buffer is freed.
• goto fail in the second js_array_sort loop, because the value array is already freed there.
• return -1 in JS_FlattenIntoArray.
• goto exception everywhere else.

Tests: eight negative tests next to the existing ones in tests/bug1672/, for join, reverse, copyWithin, slice, splice, concat, sort and flat. They have the same form as the existing ones: a one-liner on {length: 2**32-1} with the qjs:set-interrupt-handler flag. Without this change each of them runs for minutes; with it each finishes in well under a second.

fill and Array.from get no test of this kind. They allocate per element and run-test262 sets no memory limit, so a regression would show up as unbounded memory growth rather than a hang. Under a memory limit, the out-of-memory InternalError would make such a test pass anyway.

run-test262 -c tests.conf reports no errors with the change, and it builds without new warnings under the CMake warning flags.

Fixes #1753