fix(parse): require both closing brackets on an array-of-tables header

[MFA-G]

Fixes #65.

Problem

An array-of-tables header closed by a single ] is accepted when it is the last thing in the document, and one trailing character is silently discarded:

parse('[[a]')   // => { a: [ {} ] }   expected: throw
parse('[[a]x')  // => { a: [ {} ] }   expected: throw ('x' is dropped)

TOML 1.1.0 requires array-table-close = ws %x5D.5D, so [[a] cannot derive from array-table. BurntSushi/toml, CPython tomllib and Rust toml all reject it.

Cause

let k = parseKey(ctx, ']')      // consumes through the first ']'
if (isTableArray) {
    if (toml.charCodeAt(ctx.p - 1) !== 0x5d /* ] */) { throw ... }
    ctx.p++
}

ctx.p - 1 is the ] that parseKey has already consumed, so the comparison is always true and the guard can never fire. The second bracket is never required, and ctx.p++ advances past whatever happens to be there. That accounts for the whole table: at EOF the increment is a no-op ([[a]), with exactly one trailing character it swallows it ([[a]x), and with more the leftovers produce an unrelated error ([[a]xyz).

Fix

The old comparison is not dead weight — parseKey calls skipVoid(ctx, true, true) after the key, so ctx.p - 1 is what rejects tokens between the two brackets ([[uwu] ], covered by an existing test). The new condition keeps it and adds the missing requirement that ctx.p is the second bracket:

if (toml.charCodeAt(ctx.p - 1) !== 0x5d || toml.charCodeAt(ctx.p) !== 0x5d) {

charCodeAt past the end returns NaN, so EOF fails the check without a separate length test. The error pointer moves from ctx.p - 1 (the consumed bracket) to ctx.p (where the missing one should be).

Validation

• Existing suite: 141 passed, 1 skipped, no changes to the existing expectations.
• Added a case next to rejects invalid arrays of table covering [[uwu], [[uwu]x, [[uwu] and [[uwu]\nowo = 1.
• toml-test 1.1.0 with the skips from run-toml-test.bash: 690 → 692 passing. The two newly passing cases are invalid/table/array-no-close-05 ([[a]) and -06 ([[a]x), which landed in toml-test via Add invalid array-of-tables cases at end-of-input toml-lang/toml-test#205. The remaining failures (linetab-number-01..04, integer/long, utf8-bom-01/02) are pre-existing and untouched.
• Positive cases still parse: [[a]], [[ a ]], [[a.b]], [[a]] # c, [[a]]\n[[a]], CRLF variants.

Cost

One extra charCodeAt on the array-of-tables header path only — not on key-value lines, values, or plain [table] headers.

[cyyynthia]

AI garbage is not welcome here. And the fix isn't even done right but that's to be expected from a useless clanker.