Escapes a backslash inside a quoted object key

[johnnyt]

Closes px-0tz.

The defect

format_object_key/1 in StringVisitor escaped the quote character of the
style it was writing but never the escape character itself - the same defect
px-v3b fixed for string literals one screen above it, in the same visitor, and
deliberately left in place here because a quoted object key sits outside that
bead's acceptance criteria.

A quoted key is read by the same lexer string rule as a literal
(parse_object_key/1 in parser.ex consumes a STRING token), so it always
owed the same escaping. Before this branch, on main:

• {'a\b': 1} rendered with the backslash unescaped, so parsing it back
  yielded the key ab - a silent data loss with no warning.
• A key of a single backslash rendered {'\': 1}, an unterminated literal
  that does not parse at all.
• A backslash immediately before the quote rendered source whose closing
  quote landed one character early.

The fix

Both quoted clauses now route through escape_within/2, the shared helper
px-v3b introduced, rather than growing a second implementation. Two escape
paths in one writer is how this defect outlived the first fix, so the bead
asked for the helper specifically.

The :identifier clause is untouched. The parser only ever produces it for a
bare identifier, which by construction carries neither a quote nor a
backslash, and quoting one on the writer's own initiative would switch a style
the writer is required to render as asked for (the px-v3b operator ruling).

Tests

test/predicator/visitors/string_visitor_object_key_escape_test.exs pins the
round trip for a corpus enumerated from the characters the lexer treats
specially, not transcribed from examples: 129 keys x 2 quote styles = 258
cases, plus the named cases the bead calls out and a guard that the identifier
style still renders bare.

Confirmed red before the fix (4 failures, including both the lone-backslash
and backslash-before-quote cases) and green after.

Provenance

Two files the bead did not name were forced by the change and are included:

• test/test_helper.exs and
  test/predicator/visitors/string_visitor_escape_test.exs. The new suite
  needs the same awkward-value corpus as the px-v3b literal suite, and a
  second copy of the enumeration tripped Credo.Check.Design.DuplicatedCode
  on the full gate. The corpus moves to Predicator.EscapeCorpus in
  test_helper.exs - this repo's established test-support pattern, alongside
  Predicator.SpanSlicing and Predicator.ASTShape - and both suites read
  it. No behavior in either suite changed; the literal suite runs over the
  same values it did before.

No new public surface. No ISA movement, no corpus regeneration, no version
bump.

Changelog

changelog.d/px-0tz.md, a Fixed entry: user-visible, since a caller
decompiling an object with a quoted key was losing data.

Gate

mix quality green on this HEAD (rebase was a no-op; main is d952dd0):
format, compile with warnings-as-errors, deps, Credo clean, 2,895 tests at
95.5% coverage, Dialyzer clean.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
lib/predicator/visitors/string_visitor.ex	95.77% <100.00%> (ø)

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.