[helly25]

Fix the tools test when bashtest is built as an external module, as BCR does, and exercise the framework suite from an external consumer in CI. Derive consumer versions from the checkout and automate the README version update during release preparation, keeping the installation example on the published release during development.

AG;DR

• Bring the test-only correction from mboworks_bashtest@0.7.0 bazelbuild/bazel-central-registry#11164 into upstream source. Compare $(location) against Bazel's $(rootpath) instead of a hardcoded root-workspace path, preserving the existing contract. No runtime API or published 0.7.0 artifact changes.
• Extend external_consumer_test.sh to run @framework//bashtest/tests/... alongside the sample consumer with manifest-only runfiles. Existing CI invokes this script on Linux, macOS, and Windows, using a renamed repository and a workspace path containing spaces.
• Advance MODULE.bazel and CHANGELOG.md to development version 0.7.1. Both external consumer scripts derive their dependency version from MODULE.bazel because their local overrides consume this checkout. Correct the stale README example to published version 0.7.0.
• Add python3 tools/prepare_release.py X.Y.Z: validate the module/changelog and update the README to the intended release. Review and merge that preparation before tagging. The tag helper, including dry runs, and the archive workflow both require --check to pass. This preserves agreement in the tagged source, archive, and release notes without mutating published artifacts. Preparation does not commit, tag, or publish.
• The pre-commit version gate permits the latest numeric release tag in README during development, or the module version during release preparation. Thus the example briefly names the upcoming version in the release-preparation commit. Without local tags the checker can validate only the README format and upper bound; CI fetches all tags. Historical changelog entries remain unchanged, and release assets/BCR templates derive their versions from the release tag.
• Regression tests cover version mismatches, missing/ambiguous examples, invalid versions, numeric tag ordering, release-to-development transitions, preparation idempotence and rejection without writes, and actual archive generation: stale README fails before an archive exists; prepared README, module, VERSION, and notes all agree.
• Validation for 539ac74: all 44 Python tests, both external consumer scripts, bazel test //... (8 applicable tests; Windows runtime test incompatible on macOS), and pre-commit run --all-files passed. All PR CI checks passed on this head: Linux, macOS, Windows, pre-commit, release-site tests, and the final gate (run 39070227451).
• Earlier regression evidence: the extended consumer script failed on the original assertion, then passed after the correction. Previous head a311b58 passed all CI jobs, including Windows. An earlier attempt on 4e7d226 reported a missing runtime_test.exe from a cached Windows build; the retry and subsequent head passed. Its cause remains unproven; no cache or validation gates were weakened.
• Targets main with no unmerged PR dependency or required merge order. The same source correction passed all 18 BCR platform/Bazel combinations as a registry patch. This PR does not publish a release.