Fix base58 encoding losing leading zero bytes

[sumanjeet0012]

Fixes #35

This PR fixes the issue where BaseStringConverter silently drops leading \x00 bytes during encoding and decoding for integer-based encodings (like base58btc). It now properly counts and preserves them by utilizing the encoding alphabet's zero character.

Changes made:

• Modified BaseStringConverter.encode() and decode() to manually handle leading zero bytes.
• Updated Base16StringConverter.decode() to use bytes.fromhex() directly to bypass the zero-counting logic, as base16 operates byte-by-byte.
• Added comprehensive tests for base58btc, base58flickr, base32z, base36, base10, base8, and base2 with leading zero byte strings to ensure correct round-tripping.

All 287 tests pass successfully.

[acul71]

Review (maintainer) — Request changes

Thanks for tackling #35. The diagnosis is right, and non-all-zero leading-zero payloads round-trip correctly. I’m still requesting changes and prefer overlapping PR #55 for landing this fix.

Prefer #55

• #55 fixes the same BaseStringConverter leading-zero bug, handles all-zeros and empty inputs, adds tests/test_roundtrip.py across all encodings, and has green CI.
• #48 currently has no CI checks reported on fix-issue-35-leading-zeros.
• Please treat #55 as the preferred merge vehicle. Leaving this PR open (not closing) for reference / coordination.

Blockers

1. Missing newsfragment — add newsfragments/35.bugfix.rst (user-facing ReST + trailing newline). Mandatory for approval on this track.
2. All-zeros off-by-one — super().encode(0) already emits the zero digit, then leading zeros are prepended again. Locally: encode("base58btc", b"\x00") → z11 (expected z1); decode yields an extra \x00. Same pattern for other integer bases.
3. Empty input — encode("base58btc", b"") → z1 / decodes to b"\x00". Needs an empty-input special case (as in #55).

Other notes

• Branch is 6 behind / 1 ahead of origin/master; dry-run merge was clean (no conflicts) but should be synced if this PR continues.
• New fixtures only cover \x00\x00 + non-zero payload; they don’t catch the all-zeros/empty bugs.
• Local make lint, typecheck, test (287 passed), docs-ci all passed on the PR worktree — but CI absence on GitHub remains a concern.
• Minor: parameter name bytes shadows the builtin; hasattr(..., "decode") is weaker than #55’s isinstance + rename.

Bottom line: Request changes — missing newsfragment, incomplete edge cases, no CI, superseded by stronger #55. Do not merge #48 as-is; prefer #55.