feat: support arithmetic expressions in posting amounts

[0Xgit]

PR: support arithmetic expressions in posting amounts

Fixes #2648.

What this does

Posting amounts can now contain simple arithmetic, so the parts stay visible
in the journal instead of being hand-added first:

2026-01-01 shop
  expenses:groceries   $21.60 + $27.68
  assets:cash

Supported: + - * / with normal precedence, and parentheses.
$21.60 + $27.68 prints as $49.28.
The issue's example works: $-50.20 + $-209.00 + $-16.54 + $-438 gives $-713.74.
Tax-style scaling works too: (1 + 0.05) * $47.97 gives $50.3685.

Rules

• + and - need the same commodity on both sides. A bare number takes the
  other side's commodity, so $5 + 3 is $8.
• * and / need a plain number on the right. * also allows a bare number
  (or bracketed expression) on the left, so (1 + 0.05) * $47.97 is valid.
• Anything else (mixed commodities, divide by zero) is a parse error pointing
  at the bad region.
• The kept operand's display style is reused; display precision follows the
  result ($4.20 * 2 stays $8.40). Division is computed with 6 extra places,
  then trailing zeroes are trimmed.
• ) is no longer accepted as an unquoted commodity character, so bracketed
  expressions parse. Use ")" quoted if a symbol really needs it.

Files

• hledger-lib/Hledger/Read/Common.hs — expression parser (simpleamountp
  now parses additive/multiplicative/primary levels), applyArithOp, tests.
• hledger-lib/Hledger/Data/Amount.hs — one-char change to the commodity list.
• hledger/hledger.m4.md — short docs section.

Testing

• cabal run hledger-lib:unittest -- -p 'amountp': 14/14 pass (9 new cases).
• Full cabal run hledger-lib:unittest: 249/249 pass.
• Built the hledger binary and ran print on the examples above; output
  matches, and the balanced postings still balance.
• Functional tests: moved case 5 ((1 + 1) amount) out of
  ledger-compat/hledger-unsupported.test (it asserted failure, which this
  feature intentionally changes) into journal/amounts-and-commodities.test
  as case 28, asserting the new print output. Both files pass locally
  (32/32 via shelltest).

AI disclosure (required by hledger.org/AI.html rule 4)

AI assistance: yes. Provider/model: Muse Spark (Meta, agentic coding in
opencode). Rough output: on the order of tens of thousands of tokens across
debugging and test-fix iterations. I reviewed, tested, and take responsibility
for every line (rule 7). Note: this would be my first hledger PR — flagging
per rule 6; happy to rework or hand-rewrite anything the maintainer prefers.
No OpenAI models used (rule 5).