v1.1.0

Raises the platform floor to 27 and Swift 6.4. No API changes. If you are
still on OS 26 or Swift 6.3, stay on 1.0.x: SwiftPM does not consider
platforms when resolving versions, so a package whose deployment target is 26
and that depends on from: "1.0.0" will resolve to 1.1.0 and then fail to
build. Pin with .upToNextMinor(from: "1.0.6").

Also fixes three CI gates that could not fail on what they were added to catch
(see Fixed), and cleans up dead code, tools, and docs.

Changed

• The platform floor is now macOS / iOS / tvOS / watchOS 27 and Swift 6.4
  (swift-tools-version: 6.4), following macOS 27 and Swift 6.4 going GA on
  2026-09-15.
• ZPLKitVerifier is still unavailable on watchOS. The earlier stated reason
  (Vision's Swift API needs watchOS 27) was incomplete: watchOS 27 has
  DetectBarcodesRequest but its SDK has no RecognizeTextRequest at all.
• CI: the floor, cross-platform, and docs jobs moved from the hosted
  macos-26 / Xcode 26.6 image to the hosted xcode-27 image, pinned to
  Xcode 27.0. visionOS joined the cross-platform build matrix. The Linux and
  WebAssembly jobs moved to the swift:6.4 containers and the 6.4.0 Wasm SDK
  (both still run 184 tests). The test-output check that was copied into
  three jobs now lives in Scripts/check-test-output.sh.
• CI: the per-module coverage gate moved from the hosted floor job to the
  self-hosted build-and-test job. GitHub's hosted macOS 27 VM cannot run
  Vision text recognition (e5rt_e5_compiler_compile call failed on every
  RecognizeTextRequest), so the hosted job sets ZPLKIT_SKIP_VISION_OCR=1
  to skip the two suites that need OCR. Every other environment, including a
  plain local swift test, runs them.

Removed

• Dead conditional-compilation branches in Apple-only targets: the
  non-Compression fallback in the ^GF parser and a Glibc import in
  ZPLKitPrinter, neither of which could ever compile.

Fixed

• RenderFixtures exits non-zero when any fixture fails to render.

• DitherTestPrint defaulted both printers to the same host, which also
  skipped the pause between the two jobs.

• Docs: the PrinterCommand example called an API that does not exist; ~JA
  is now consistently described as cancelling all jobs; IntelligentMail is
  listed under 1D barcodes in the DocC topics; CONTRIBUTING no longer describes
  Xcode 27 beta bugs or a macOS test-count floor that no longer exist.

• The coverage gate could not fail on a module that vanished from the
  report. Scripts/coverage.sh looped over the modules llvm-cov reported,
  never over its own floor list, and 1.0.6 justified the gate on the premise
  that a dropped test target takes its module's coverage toward zero. That
  premise is wrong: ZPLKitPrinter is linked only by ZPLKitPrinterTests, so
  when that bundle stops running the module does not sink, it leaves the report
  altogether. The script then printed "All modules at or above their coverage
  floors" and exited 0. Since the test-count assertion had been relaxed to
  "greater than zero" in the same release, a silently dropped
  ZPLKitPrinterTests passed every job. The gate now iterates the floor list,
  and a module named there and missing from the report is a hard failure.
  Verified by removing the built ZPLKitPrinterTests bundle: previously green,
  now fails.

• Module names no longer depend on llvm-cov's path shortening. The script
  read the module from the first path component of the text report's filename
  column, but llvm-cov report strips the longest common path prefix. Over the
  whole suite the rows read ZPLKit/Elements/Aztec.swift; over a single target
  they read Aztec.swift, every module became unrecognised, and the run passed
  with nothing enforced. It now reads llvm-cov export -summary-only, whose
  JSON carries absolute paths. Reported numbers are unchanged (86.77% line,
  80.25% region).

• A failing test under WebAssembly now fails CI. The Wasm job treated any
  non-zero swift test exit as "this toolchain cannot host tests yet",
  printed a notice and passed, including for a genuine test failure or a
  runtime trap, which is the class of regression the step exists to catch. It
  now degrades to a notice only when the test runner produced no output at all
  (an SDK or link failure, which never reaches the runner) and fails the job
  when tests ran and failed.

• A test-count step that found no count now says so. Under pipefail, a
  no-match grep in the counting pipeline aborted the step before its own
  error message could print, so "nothing ran" failed the job with no
  diagnostic.