Fix the dollar amount parser behind a passing test

An AI added a unit test for your ParseAmount function, which turns dollar strings like $12.34 into cents. The test passed, but it only ever tried well-formed amounts, so it could not fail. Your team's review writes down the parser's full contract and checks the code against it.

An assistant wrote parse_amount_cents(s). It is in the editor. It has a bug. Find it and fix it.

Input and output:

• s: the amount as text.
• Return the amount in cents as an integer written as a string, with no leading zeros and a leading - when it is negative, or "error".

Rules the tests check:

• An optional - at the very start makes the amount negative. A $ must come next. Without it, return "error".
• After the $ comes the dollar part, then optionally a . and exactly 2 digits of cents. With no ., the cents are 00. Any other number of characters after the ., or anything but digits 0 to 9 there, is an error.
• The dollar part is one or more digits 0 to 9, which may be grouped with commas as thousands separators. With commas, the first group has 1 to 3 digits and every later group has exactly 3. An empty group, a group of the wrong size or any other character is an error.
• Leading zeros in the dollar part are allowed.
• Minus zero is written "0".