node:buffer: default lastIndexOf(value, encoding) to search from the end (#31507)

`Buffer.prototype.lastIndexOf(value, encoding)` — the documented 2-arg
overload where `byteOffset` is omitted and the second positional
argument is the encoding — searched from offset `0` instead of from the
end, so it returned `-1` for substrings that are actually present (or
the wrong index). Broken for every encoding passed as the 2nd arg
(utf8/latin1/hex/utf16le/base64).

`JSBuffer.cpp` hardcoded `byteOffsetD = 0` in the
`byteOffsetValue.isString()` branch, which defeated the direction-aware
default a few lines below (`if (isnan(byteOffsetD)) byteOffsetD = dir ?
0 : byteLength`). Setting it to `NaN` lets that default apply:
`lastIndexOf` now searches backward from `byteLength` (matching Node),
while forward `indexOf`/`includes` (which default to 0 anyway) are
unchanged, as is the 3-arg `lastIndexOf(value, byteOffset, encoding)`
form.

Adds tests covering `lastIndexOf(value, encoding)` across
utf8/latin1/hex/utf16le/base64. (This matches Node — Bun 1.3.14 had the
same bug, so it's not a port regression.)

---------

Co-authored-by: robobun <117481402+robobun@users.noreply.github.com>