Remove dangling references and environment-specific paths

[GNERSIS]

An audit of the tracked tree turned up references that either cannot resolve on a clean clone or assume one machine's environment. Companion PRs exist for PJ4 and pj-official-plugins.

This is the public SDK, so a reference that only resolves on a maintainer's machine reaches the widest audience of the three repositories — an external contributor hits it on their first git clone.

Scope rule applied: fix only what breaks or exposes a machine. Genuine author/copyright attribution is deliberately untouched — the 329 Davide Faconti copyright sites, LICENSE, LICENSE-APACHE and the upstream dependency URLs are all unchanged.

Broken on a clean clone

Cited from	Problem
README.md:38 and .claude/commands/test-and-commit.md:5	Both instruct the reader to run ./run_clang_tidy.sh, which is not tracked. It is step 3 of the documented getting-started flow, so every external contributor following the README hits it. Replaced with the equivalent git ls-files … | xargs clang-tidy command in both places, kept in sync.
CMakeLists.txt:58	Cites SUSTAINABILITY.md T0-3 for the wasm -Werror carve-out; no such file is tracked anywhere. The adjacent comment already explains the reason, so the dangling cross-reference is dropped and the comment stands on its own.

Environment-specific

• pj_plugins/tests/source_dialog_integration_test.cpp:200 used 10.0.0.1 — RFC 1918 space that can overlap a contributor's real LAN. Now 192.0.2.1 (TEST-NET-1), the documentation-only equivalent, with identical test semantics.
• pj_plugins/docs/dialog-plugin-guide.md:776 offered a copyable example with initial_directory = "/data/bags", a root-level mount an arbitrary machine will not have. Now {}, letting the host pick.
• .github/workflows/release.yml wrote /tmp/remote_revs.json in five places rather than the runner-allocated temp dir, bypassing GitHub's temp-path abstraction and giving concurrent or local runs a fixed collision target. Now one mktemp under $RUNNER_TEMP with a cleanup trap, quoted at all five uses.

Verification

• release.yml re-parsed as YAML after the edit.
• clang-format --dry-run -Werror clean on the changed test.
• Not compiled. The only C++ change is a string literal in a test; the rest is docs, CMake comments and CI shell.

Deliberately not fixed

pj_base/abi/baseline.abi embeds /home/runner/work/plotjuggler_core/... on lines 1 and 100 — CI checkout paths baked into a distributed artifact. Regenerating it properly means re-running abidw with --no-corpus-path --no-comp-dir-path --short-locs (or a -ffile-prefix-map build) and diffing the result against the current ABI contract. That needs a build, so hand-editing the baseline would be worse than leaving it. Flagged for a follow-up.

Also left alone under the scope rule, since neither breaks nor exposes a machine: CHANGELOG.md:748,766 names a person for the pj.data_processors.v1 contract, and pj_plugins/CMakeLists.txt:677 uses a "Keystone" codename that appears exactly once and is defined nowhere.

https://claude.ai/code/session_01JHCjCjntYefq2QyDagb8Dh