base64: Support multi bytes encoding characters

[randax]

Summary

• encode()/decode() in @xmpp/base64 use btoa()/atob(), which only handle Latin-1 (single-byte) encoding
• Non-ASCII characters like ø (U+00F8) get encoded as a single byte (0xF8) instead of the correct UTF-8 sequence (0xC3 0xB8)
• This silently corrupts SASL PLAIN credentials for users with non-ASCII usernames, causing authentication failures downstream

The fix replaces btoa()/atob() with TextEncoder/TextDecoder to properly handle UTF-8, and adds a test suite covering Nordic (æøå), German (äöüß), French (éèêëç), Cyrillic, CJK, Arabic, Hebrew, and emoji characters.

Reproduction

Any username containing characters outside ASCII will produce corrupted SASL PLAIN payloads. For example, with username øyvindranda:

import { encode } from "@xmpp/base64";

// Current (broken): btoa encodes ø as single byte 0xF8
encode("ø"); // => "+A==" (base64 of [0xF8])

// Fixed: TextEncoder encodes ø as UTF-8 bytes 0xC3, 0xB8
encode("ø"); // => "w7g=" (base64 of [0xC3, 0xB8])

When a server decodes the +A== payload, byte 0xF8 is not valid UTF-8, so the username gets mangled (e.g. replaced with U+FFFD), and JID construction or session binding fails silently.

Test plan

• All 18 new tests in packages/base64/test/test.js pass
• Full test suite passes (201 passed, 1 skipped, 0 failed)
• ASCII-only credentials produce identical base64 output (backwards compatible)
• Verified with real XMPP server: Nordic username øyvindranda authenticates successfully after fix

🤖 Generated with Claude Code

[sonnyp]

Thanks

I slightly improved the impl a7279c4

[sonnyp]

The new tests are passing, CI failures are unrelated and will be addressed separately.