refactor: use token-based terminator detection in parse_command

[ldayton]

Summary

• Add _lex_is_command_terminator() method that uses the Lexer to detect command terminators via token types
• Update parse_command() to use token-based terminator detection instead of character-level checks
• Remove the now-unused _is_list_terminator() function
• Update PLAN.md to document the hybrid approach progress

Motivation

This change improves the Parser's architectural alignment with bash's tokenizer-parser separation. By using token-based terminator detection, the EOF token mechanism (implemented in #327) now works at the command level, not just the list level.

Key Insight

The hybrid approach works because:

• _lex_peek_token() doesn't advance Parser position (only _lex_next_token() does)
• This allows token-based terminator checks followed by character-based word parsing
• Position syncing via _sync_lexer() bridges the two modes

Changes

# Before (character-based):
if self.at_end():
    break
ch = self.peek()
if _is_list_terminator(ch):
    break
if ch == "&" and not (self.pos + 1 < self.length and self.source[self.pos + 1] == ">"):
    break

# After (token-based):
if self._lex_is_command_terminator():
    break

All 4515 tests pass in both Python and JavaScript.