CommandParser: keep a comma inside a KEY=VALUE argument (#120)

[vv137]

Fixes #120.

The bug

:align mob to ref chain=A,B never applied the chain filter.

The command tokenizer treats , as an argument separator identical to
whitespace, so the modifier arrived as chain=A plus an orphaned B.
popModeToken pops recognised modifiers off the back of the argument list, hit
the bare B first, and stopped — leaving chain1/chain2 unset and the
remnants to leak into the target selection:

$ :align 1ubq to 1brs chain1=A,B chain2=A,B
Target selection empty: chain1=A B chain2=A B

All four documented examples of the chain=A,B,… shorthand (issue #81) were
dead, on :align, :alignto and :loadalign alike.

The fix

No caller could have recovered this: chain=A B and chain=A,B tokenize
identically, so the information is gone before the command ever sees it. Fixed
where it is lost — a comma inside an argument that already contains =
belongs to the value.

Bare commas keep separating, so the PyMOL-style forms are untouched:

input	tokens
rename old, new	old · new
color red, obj 1ubq	red · obj · 1ubq
set bg rgb(32,32,32)	bg · rgb(32 · 32 · 32) (still rejoined by :set)
align m to r chain=A,B	m · to · r · chain=A,B

The rgb(R,G,B) path is worth calling out: none of its tokens contain an =,
so it still splits and :set still rejoins it exactly as before.

Verification

• Added tests/test_command_parser.cpp — the
  tokenizer had no test coverage at all. Pins both halves of the rule, plus
  quoting and rawArgs.
• chain=A,B now agrees with chain=A+B atom-for-atom; chain1=A,B chain2=A,B
  went from "Target selection empty" to a correct MM alignment.
• Manually re-checked every comma-dependent command (:rename, :color,
  :show scope forms, :set bg, :set label_color) and the 59-expression
  selection corpus — no change.
• Full suite green.

Note

Write the list without spaces. chain=A, B is still two arguments and the B
is not part of the filter — documented in the README alongside the shorthand.

🤖 Generated with Claude Code