v1.1.1

A documentation-accuracy patch. No runtime behaviour changed — every command,
flag, and MCP tool works exactly as in 1.1.0. What changed is that the help text
now tells the truth about which providers you can pass.

Fixed

• target list --provider advertised only two of the four providers.
  kuberoutectl target list --help said filter by provider (azure|aws) while
  gcp and kubeconfig were registered and the filter accepted them — it passes
  the value straight through, there was never a whitelist. Only the text was
  wrong, which is why nothing failed when it drifted. It was stale in two places:
  the flag usage and the command's long description. (#104)

  The list is now derived from the provider registry instead of written by hand,
  so it cannot go stale again, and a test asserts every registered provider
  appears in the help of both target list and credential list.

  Note the ordering is now alphabetical (aws|azure|gcp|kubeconfig), since it
  comes from the registry's sorted listing.

Added

• A regression guard for the MCP tools' provider jsonschema descriptions,
  which carried the same hand-written provider list. They were already accurate —
  nothing was broken — but nothing tied them to the registry. Struct tags are
  compile-time constants and cannot be derived, so a test now reads them by
  reflection and requires the listed ids to match the registry exactly. No schema
  or tool behaviour changed. (#105)

Verified

• Reproducible builds confirmed for this tag — two consecutive
  make snapshot runs produced an identical dist/checksums.txt, covering the
  archives and the .deb / .rpm packages. This check was skipped for v1.1.0;
  from this release onward it runs before the tag is pushed.
• Unit tests, make check, and scripts/e2e.sh green on the tagged commit
  (cb6ec00) via ci.yml on main.
• The published linux_arm64 artifact was downloaded, checksum-verified, and
  run: it reports 1.1.1 (commit cb6ec00), and target list --help shows all
  four providers.

Not verified

• No third-party MCP client (Claude Desktop and friends) is exercised by the
  test suite. The stdio protocol itself is verified end to end against the
  shipped binary, but a real client integration is not.
• No provider tool runs against a real cloud account — provider flows are
  fixture-driven with fake az / aws / gcloud / kubectl.

Full changelog: https://github.com/ymedlop/kuberoutectl/blob/main/CHANGELOG.md