fix(core): preserve binary output in ProcessOutput.buffer()

[spokodev]

Problem

ProcessOutput.buffer() builds its Buffer from this.stdall:

buffer(): Buffer {
  return Buffer.from(this.stdall)
}

But stdall is a getter that UTF-8 decodes the stored chunks
(bufArrJoin → TextDecoder('utf-8').decode). Any byte that isn't valid
UTF-8 has already been replaced with U+FFFD by the time buffer() runs, so
the bytes it returns are corrupted — even though the original, lossless chunks
are still sitting in this._dto.store.stdall.

blob() and text(<encoding>) both delegate to buffer(), so they are
affected too. These are the documented APIs for getting bytes out of a command:

const png = await $`cat logo.png`.buffer()   // mangled: every non-utf8 byte → ef bf bd
await $`cat logo.png`.text('hex')            // same corruption

Reproduction

const o = await $`node -e ${'process.stdout.write(Buffer.from([255,254,129,130]))'}`
o.buffer().toString('hex')   // 'efbfbdefbfbdefbfbdefbfbd'  ← expected 'fffe8182'

Fix

Reconstruct the Buffer from the raw store chunks (which are the original
string | Buffer pieces) instead of the decoded string:

buffer(): Buffer {
  return Buffer.concat(
    [...this._dto.store.stdall].map((chunk) =>
      isString(chunk) ? Buffer.from(chunk) : chunk
    )
  )
}

This also fixes blob() and text('hex' | 'base64' | 'latin1' | …), which go
through buffer(). The plain-text path (toString() / text()) is unchanged.

Test

Added a buffer() case in test/core.test.js asserting that non-UTF-8 bytes
survive buffer()/text('hex'). Fails on main, passes with the fix; the
existing buffer()/text()/blob() string tests still pass.