refactor: use Uint8Array instead of Buffer

[diegomura]

What kind of change does this PR introduce?

Nothing under lib/ uses Node's Buffer anymore.

Latin1 and UTF-16BE encoding, hex and base64 conversion, concatenation and fixed-size allocation all go through Uint8Array and a few helpers in lib/binary.js now.

This is part of modernizing the library, and follows the fflate and base64 changes. Buffer is a Node builtin that bundlers no longer polyfill by default, and it was the last one left in code that has to run in a browser.

It is not a breaking change. Buffer is a Uint8Array subclass, so passing one to image(), file() or registerFont() works as before, and Readable.push() converts Uint8Array to Buffer zero-copy, so Node consumers still read Buffer chunks off the document stream.

pdfkit/virtual-fs now returns a Uint8Array from readFileSync(). Encoded reads still return strings.

A dict value that is a plain Uint8Array now serializes to a hex string. The check was Buffer.isBuffer(), so it used to fall through to String(obj) and write 1,2,3.

TextDecoder('latin1') is not a valid replacement for the String.fromCharCode loops. Node maps latin1, iso-8859-1 and windows-1252 all to ISO-8859-1, but the WHATWG Encoding Standard makes all three windows-1252, where 0x80 decodes to U+20AC. Browsers follow the spec, so it would corrupt 0x80-0x9F in the browser build.