fix(utf8): strip only one leading BOM on decode

[spokodev]

Decoding UTF-8 strips two leading BOMs instead of one, dropping a U+FEFF that belongs to the content.

For UTF-8 the internal TextDecoder is created with ignoreBOM: false, so it already removes one leading BOM. The decoder is then also wrapped by iconv-lite's StripBOM (added for every bomAware codec unless stripBOM: false), which removes a leading U+FEFF as well. When the byte stream is BOM followed by content that begins with U+FEFF, both strip and the content's U+FEFF is lost.

Repro:

const iconv = require("iconv-lite")
const buf = Buffer.concat([Buffer.from([0xEF, 0xBB, 0xBF]), Buffer.from("abc")])
iconv.decode(buf, "utf8")            // => "abc"     (U+FEFF dropped)
new TextDecoder("utf-8").decode(buf) // => "abc"

utf-16le/utf-16be/utf-32* all strip exactly one BOM here; only utf-8 strips two.

Fix: let StripBOM be the sole BOM stripper by setting ignoreBOM: true on the internal TextDecoder. The wrapper is present whenever a BOM should be removed (and absent for stripBOM: false, where the BOM should be kept), so a single strip is correct in every case. The stripBOM: false and stripBOM callback paths are unchanged.

Oracle: the WHATWG Encoding Standard / TextDecoder("utf-8") removes exactly one leading BOM.

Adds a regression test.