assert: keep the beginning of long strings when truncating

[arhxam]

Description

addEllipsis() in lib/internal/assert/assertion_error.js truncates the
long single-line actual / expected values shown by AssertionError's
custom util.inspect output. For a string longer than
kMaxLongStringLength (512) it did:

return `${StringPrototypeSlice(string, kMaxLongStringLength)}...`;

StringPrototypeSlice(string, 512) drops the first 512 characters and
keeps the tail, then appends ... as if the end had been cut off —
so the ellipsis is misleading and the wrong part of the string is shown.

Reproduction on current main:

const assert = require('node:assert');
const util = require('node:util');
let err;
try { assert.strictEqual('a'.repeat(300) + 'b'.repeat(300), 'x'); }
catch (e) { err = e; }
console.log(util.inspect(err));
// actual: 'bbbbbbbb…b...'   <- only the last 88 chars survive; no 'a' at all,
//                              and the trailing '...' implies the end was cut

The three sibling truncations in the same file and the multi-line branch
of this very function all keep the beginning of the string. This
changes the single-line branch to StringPrototypeSlice(string, 0, kMaxLongStringLength) so the first 512 characters are kept, consistent
with the trailing ....

Why the existing tests did not catch it

The existing tests truncated homogeneous strings ('A'.repeat(n)), where
the head and tail are identical, so the buggy tail length
(n - 512) was baked into the expectations (e.g. 'A'.repeat(488),
substring(0, 9_488)). Those expectations are corrected to 512, and a
new heterogeneous regression test ('a'.repeat(300) + 'b'.repeat(300)) is
added that actually distinguishes the beginning from the end.

Verification

• Reproduced the bug on a from-source build of main, then confirmed the
  fix flips it (the new test fails on unpatched source, passes with the fix).
• python3 tools/test.py on all test/parallel/test-assert*.js (14 files)
  and test-util-inspect.js: all pass.