dqcheckrGUI

Bug fixes

• The drift-comparison report link now works again. The GUI located the rendered drift report by reconstructing its filename with a fixed pattern, which stopped matching once dqcheckr's drift filenames gained a snapshot-id suffix — so every drift comparison reported "report not found" even though it rendered. The GUI now uses the report path compare_snapshots() returns directly (requires dqcheckr >= 0.2.5 behaviour), removing the fragile filename contract.
• Report links in the dataset panel now account for a report that is still rendering. A snapshot whose render_status is "pending" (dqcheckr 0.2.5+) no longer gets a clickable "Open" link to a file that is not yet on disk; it shows "rendering…" until the report is written. A filename is only reconstructed for a genuine pre-0.2.3 row, not whenever one is absent.
• The dqcheckr dependency floor was raised to the version whose behaviour the report and drift link builders rely on.
• Config-file writes are now crash-safe. Both dataset configs and the global dqcheckr.yml are written to a temporary file and renamed into place, so an interrupted write (process killed, power loss, a contended OneDrive/network share) can no longer leave the sole config for a dataset truncated and unparseable.
• A malformed global dqcheckr.yml no longer crashes the session at startup or gets silently overwritten with a trimmed copy on save. Startup falls back to the built-in defaults with a warning, and a save aborts with "Save failed" rather than wiping keys it could not read.
• Read failures are no longer disguised as empty results. A corrupt or unreachable snapshot database is reported distinctly instead of showing "No run history found", and editing a dataset whose YAML cannot be parsed now shows an explicit error and leaves the file untouched, rather than opening a blank wizard that would overwrite it on save.
• Background-process races were fixed. Clicking Stop on a run that finished in the same instant now reports its real PASS/WARN/FAIL result instead of "stopped", and a second drift comparison launched while one is still running is refused rather than orphaning the first.
• A renamed-header CSV config no longer loses its col_names/csv_skip when edited and saved without re-visiting the structure step, and the step-3 encoding sniffer no longer crashes on a large, mostly-ASCII file whose charset detector returns an indeterminate result.
• Hardened the observers behind client-settable inputs so a malformed payload can no longer error them, and added test coverage for the run panel, the global-config reader, and the dataset-name JavaScript escaper.
• The step-3 encoding sniffer no longer writes encoding: ASCII into dataset configs. readr::guess_encoding()'s head sample confidently reports "ASCII" whenever the first accented byte sits beyond the sample — and that config value later hard-crashed the run's R subprocess ("Invalid multibyte sequence" inside vroom/iconv) once a delivery contained non-ASCII bytes. Step 3 now scans the *entire* file: valid UTF-8 (which includes pure ASCII) is applied as UTF-8 with certainty; only when the file is genuinely not UTF-8 does the statistical detector run — seeded with a chunk that actually contains the non-ASCII bytes — and its legacy-encoding candidates go through the usual confirm flow. New dependency: stringi.
• An encoding: ASCII in an existing config (written by hand or by an older GUI) is normalised to UTF-8 when the config is read, so editing and re-saving a dataset upgrades it. ASCII is a strict subset of UTF-8, so the change is lossless.