Long path-like values in diagnostic details (e.g.
`/auto_stop/slow_down_mid/brake_trigger`) were rendered outside the
grid item's box and visually overlapped the next key:value pair, a bug
that has existed since diagnostics support was first introduced in #1.

Add `min-w-0` + `shrink-0` + `break-all` so long values wrap within
their column, and add an e2e test that injects a realistic long value
into the DOM and asserts `scrollWidth <= clientWidth` on each grid
item, which is the exact invariant the fix preserves.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* Add UTF-8 BOM to CSV exports and add export output tests

CSV exports lacked UTF-8 BOM, causing LibreOffice to misdetect encoding
as Shift-JIS for Japanese text. Also adds comprehensive unit tests for
all CSV/JSON/TXT export functions and E2E tests verifying downloaded
file content including BOM presence.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* Fix diagnostics detail value overflow and add overlap regression test

Long path-like values in diagnostic details (e.g.
`/auto_stop/slow_down_mid/brake_trigger`) were rendered outside the
grid item's box and visually overlapped the next key:value pair, a bug
that has existed since diagnostics support was first introduced in #1.

Add `min-w-0` + `shrink-0` + `break-all` so long values wrap within
their column, and add an e2e test that injects a realistic long value
into the DOM and asserts `scrollWidth <= clientWidth` on each grid
item, which is the exact invariant the fix preserves.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* Address Copilot PR feedback: CR in CSV quoting and explicit BOM stripping

- `escapeCSV` now also quotes values containing `\r`, so fields with
  Windows CRLF (plausible in ROS log messages) produce RFC 4180-valid
  output. Add unit tests for both lone `\r` and `\r\n`.
- Export-content E2E tests now use a `stripBomAndSplitLines` helper that
  explicitly removes a leading `\uFEFF` and splits on `/\r?\n/`, rather
  than relying on `String.prototype.trim()` to incidentally drop the BOM
  (which also strips meaningful whitespace).

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* Expand diagnostics fixture to cover 1/3/4-value and long-value cases

The test_sample.bag diagnostics previously had mostly 2-value entries,
which didn't exercise the real-world shape of diagnostic messages or
the overflow condition the recent UI fix targets.

- @0.0 /sensor/lidar OK: 2 → 4 values (full lg:grid-cols-4 row)
- @2.0 /sensor/lidar ERROR: 2 → 3 values, with one realistic long
  path-like value (`last_heartbeat_source`) that reproduces the
  grid-cell overflow case
- @3.0 /sensor/lidar STALE: 1 value (unchanged; kept as single-value case)
- WARN entries at @1.0 /sensor/camera and @3.0 /motor/left left
  untouched — they are pinned byte-for-byte by the parquet export test

Regenerated test_sample.bag via the existing ros:noetic-ros-base Docker
recipe in the script's docstring. All 104 e2e tests still pass.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* Sync mcap fixture diag_data with bag fixture

The mcap generator had a comment saying "Same data as
generate_test_bag.py for consistency" but the diag_data was stale
after the previous commit expanded the bag fixture to cover 1/3/4-value
and long-value cases.

Apply the same diag_data changes and regenerate test_sample.mcap,
test_sample.mcap.zstd, and test_sample_truncated.mcap via the existing
Dockerfile.mcap recipe. All 104 e2e tests pass.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* Use full-width row for long diagnostic key-value pairs

Long values like `/rover/sensors/lidar_front_driver/heartbeat_monitor_node`
were being crammed into a narrow grid column with `break-all`, splitting
the path at every character boundary and making it very hard to read.

Now key-value pairs whose combined length exceeds 40 characters get
`col-span-full`, so they take an entire grid row and the value stays
readable without forced character-level breaks. Also switch from
`break-all` to `break-words` — only breaks when the value actually
overflows, producing more natural wrap points.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* Address Copilot feedback: update stale comments and strengthen CSV level assertion

- Update e2e test comments to reflect current implementation
  (`break-words` + `col-span-full`, not `break-all`; fixture now
  includes long values)
- Strengthen `exportDiagnosticsToCSV` level-name test to assert on the
  Level column (index 3) specifically via `split(',')[3]` instead of
  `toContain()` on the full line, which could match 'OK' in the
  message field ('OK running')
- Update PR description to match the current approach

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

---------

Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]>