Docs: CATALOG.md never mentions `curate --sql-file` and lacks a catalog-root troubleshooting entry

[kstonekuan]

Two small gaps remain in docs/CATALOG.md: it does not explain a real
curate flag, and it has no troubleshooting entry for a common catalog-root
mistake.

Current behavior

curate --sql-file is undocumented

The positional SQL argument and --sql-file flag are defined at
src/hflow/cli.py:81-90. Passing both or neither prints
curate: pass exactly one of a SQL string or --sql-file and returns 2 at
src/hflow/cli.py:676-679.

The command-line example at docs/CATALOG.md:80-85 shows only inline SQL.
Meanwhile, examples/egocentric/README.md:79-83 uses --sql-file, so a reader
can encounter the flag in an example before the reference explains it.

Document --sql-file and the exactly-one rule in the command-line section of
docs/CATALOG.md.

The catalog-root error has no troubleshooting entry

src/hflow/curation.py:136-143 raises <location> is not a catalog root when
--catalog points at a data root instead of the catalog created beneath it.

Add a short troubleshooting entry explaining that the command expects the
catalog location, commonly <data_root>/catalog, rather than <data_root>.

Pattern to copy

Use the existing command-line block at docs/CATALOG.md:80-85 and the concise
troubleshooting style of the "Port 8080 is taken" entry at
docs/RUNTIME.md:407-412.

Definition of done

• docs/CATALOG.md documents --sql-file and the exactly-one rule.
• docs/CATALOG.md explains what is not a catalog root means and what path to
  pass instead.

Validation

lychee --no-progress --include-fragments \
  --exclude '^https://github\.com/Hebbian-Robotics/hflow/(issues|security/advisories/new)$' \
  --exclude-path references/mcap-spec.md \
  --exclude-path references/foxglove-CompressedVideo.proto .

[promiseeuler]

While checking current main at exact commit 2f99071, I confirmed both documentation gaps still exist: the CLI defines --sql-file and enforces exactly one of inline SQL or --sql-file, while docs/CATALOG.md documents only the inline form; and the catalog loader’s current error explicitly points users from a data root to the catalog location (for example, <data_root>/catalog), but CATALOG.md has no troubleshooting entry for it.

I’d like to take this. I’ll keep the change documentation-only and limited to those two items, then run the repository’s prescribed Lychee command and report the exact result.

[kstonekuan]

Assigned, and thank you for re-verifying both halves against an exact commit rather than the issue text.

One thing I owe you now rather than later: the good first issue pool is best kept for people making their first contribution, and with #106 and #117 merged you are past that. Nothing wrong with taking this one, and I would rather you had it than have it sit. But for the next one, the higher-leverage work is where you have already shown range: you filed #108 and #111 with reproductions and correct root causes, and you found a real error in my #97 spec that nobody else caught.

Two live examples, neither a starter issue: #96 (the transform stamps a mid-GOP pass-through stream as conforming while doctor rejects it, and you have already done the ordering analysis it needs) and #97 (slerp, where you have already corrected the test spec). Both unclaimed. Never a rule, always an invitation.