feat: reversible SLP1 anunāsika (m̐ → M~) vs anusvāra (ṃ → M)

[Yuzki]

Summary

ISO 15919 distinguishes anunāsika m̐ from anusvāra ṃ, but standard SLP1 uses a
single token M for both, so ISO → SLP1 → ISO lost the distinction. The m̐
row now emits a project-specific extended SLP1 token M~, while ṃ keeps plain
M, making the pair reversible.

Contract

Input	From → To	Output
am̐	iso → slp1	aM~
aṃ	iso → slp1	aM
aM~	slp1 → iso	am̐
aM	slp1 → iso	aṃ

• Reverse M → ṃ is preserved by the existing first-win rule (the ṃ row
  precedes the m̐ row).
• M~ → m̐ works because the matcher sorts source tokens longest-first, so the
  2-char M~ is matched before bare M.
• M~ is a deliberate, reversible, non-standard SLP1 extension. Consumers
  needing strict external SLP1 should strip a trailing ~ (documented in README).

Changes

• SanskritTransliteration/transliteration_table.csv — m̐ row SLP1 column
  M → M~; all other rows/columns unchanged.
• tests/test_transliteration.py — new TestReversibleAnunasika (6 tests:
  4 contract cases + 2 ISO round-trips).
• README.md — new "SLP1 Anunāsika Extension" section.

Verification

• uv run pytest → 20 passed
• uv run ruff check / uv run ruff format --check → clean
• uv run ty check SanskritTransliteration tests → clean

Closes #2

🤖 Generated with Claude Code