Selection: accept , and whitespace as list separators (#116)

[vv137]

Fixes #116.

The bug

resi 2,3,6 silently selected only residue 2 — no warning, no error, exit 0.

The comma never reaches the selection grammar: CommandParser::parse splits
argv on commas as well as whitespace, so :color red resi 2,3,6 arrives at the
parser as the string "resi 2 3 6". The grammar only accepted + between list
entries, so it read resi 2 and dropped the rest without a word.

That is the dangerous shape the issue calls out — a figure coloured from a
truncated selection looks entirely plausible and survives visual review.

The fix

Value lists now treat +, , and plain whitespace as the same separator:

:color red resi 2+3+6     " 26 atoms
:color red resi 2,3,6     " 26 atoms  (was 9 — resi 2 only)
:color red resi 2 3 6     " 26 atoms

Two prongs, because the comma survives on some paths and not others:

• The tokenizer maps , to the + token, which covers commands that pass the
  raw argument tail through (:select, pca(...)/centroid(...)).
• Value lists also accept bare adjacency, which is what a comma degrades into
  after CommandParser. A list ends at an operator or the next primary
  keyword, so chain A,B and name CA still parses as it reads, and - keeps
  its range meaning (resi 10-12,20 = 10, 11, 12, 20).

Also fixed, same shape

• resn and element accepted a single value only — not even + lists.
• imgt's numeric set stopped at the first non-+ token; its region names
  truncated on every separator (imgt cdr1,cdr2 was cdr1). An unknown region
  now voids the whole term rather than quietly narrowing to the ones that
  parsed.
• imgt was missing from isPrimaryKeyword, so it neither terminated a list
  nor bounded a :loadalign file-glob tail.
• Slash components split on a gap too: /1brs/A/10,11/ is 10+11, not residue
  1011.
• docs/SCRIPTING.md promised pca(resi 1,2,3) == pca(resi 1 2 3). Both were
  truncating; both now work, and the comma→space pre-pass in ExprContext that
  existed to fake it is gone.

Verification

• 59 pre-existing selection expressions (operators, slash forms, within/same
  /byres, object qualifiers, IMGT, nucleic selectors) produce byte-identical
  results before and after.
• New unit tests assert +/,/space equivalence across all six keywords and
  both slash positions, and pin the boundaries: a list stops at an operator and
  at a primary keyword, and adjacency never overrides an explicit operator.
• Full suite green (264 selection assertions).

Deliberately not in this PR

• The general silence. Selection::parse still ignores unconsumed trailing
  tokens, so other malformed expressions stay quiet. Making that loud needs an
  error channel on Selection::parse and its ~16 call sites — filed separately.
• :align chain=A,B. Same root cause one level up, but unreachable from the
  grammar: chain=A and B are separate argv tokens before any selection
  parsing happens. Filed separately.

🤖 Generated with Claude Code