Tests: Snowflake and Databricks error / lifecycle / edge paths via wiremock

[rollroyces]

Summary

Adds nine wiremock tests across the two HTTP-family query engines
for error and lifecycle paths that current dev does not already cover.

crates/utopia-server/src/query_engine/snowflake.rs:

• auth_failure_body_is_surfaced — 401 / 403 / 503 share the same
  Snowflake {status}: {message} surfacing path; one body template
  runs three of them so a future unification or refactor of the
  error-mapping gets caught immediately.
• non_json_error_body_falls_back_to_plain_text — 5xx with a
  non-JSON body (e.g. Internal error: warehouse suspended) still
  reaches the caller as a printable message; this pins the
  serde_json::from_str::<StatementResponse> failure branch in
  run.
• business_error_code_in_a_2xx_body_is_bubbled_up — a successful
  2xx response carrying code: 000605 ("Statement was canceled by
  user") with no meta triggers the code != "090001" branch in
  run and surfaces the message; this is the "looks OK but isn't"
  case that would otherwise pass through as if nothing failed.

crates/utopia-server/src/query_engine/databricks.rs:

• an_unauthorized_http_status_is_surfaced_as_a_connection_error —
  401 surfaces through reqwest::error_for_status(). Pinned so
  callers can recognise auth failures by status code, not message.
• a_5xx_response_is_surfaced_with_status_code — 503 same shape.
• a_canceled_statement_carries_a_distinct_message — state: CANCELED
  is a separate lifecycle state from FAILED, with error_code: CANCELED;
  the message must reach callers distinct from "the query failed".
• pending_state_without_statement_id_is_surfaced — state: PENDING
  without statement_id would loop forever pulling non-existent handles.
  The existing ok_or_else("Databricks returned no statement_id")
  bail needs this as a regression anchor in case anyone refactors the
  id.required() check.
• a_succeeded_statement_with_no_rows_returns_empty — SUCCEEDED
  with an empty data_array returns rows.is_empty(); pins the
  "DDL / 0-row query / empty schema" path through run.

Why v2

The previous attempt (#1207) was based on a fork c9760a1 that had
diverged from upstream/dev (now 25fa9f4, rc9 + ~20 commits).
crates/utopia-server/src/query_engine/snowflake.rs and
.../databricks.rs have both been substantively rewritten in upstream
since — Snowflake gained result-partition reading (#1062), Databricks
gained follow-chunk-links (#1063) and information-schema fallback
(#241, #702). The wiremock body shapes used by these tests were
chosen to drive the early-return paths in run (non-success status
code, missing handle, FAILED/CANCELED state) and so the test bodies
do not need to change with the new code. They are simply re-applied
to the current tree.

What this PR is not

• No new engine. Snowflake and Databricks engine bodies are untouched.
• No behavior change to existing tests. The existing
  a_synchronous_answer_is_typed_by_row_type,
  a_202_is_polled_and_its_partitions_keep_the_first_rows_types,
  an_error_body_is_surfaced,
  a_schema_read_follows_partitions_until_exhausted,
  multiple_partitions_require_a_statement_handle,
  failed_or_incomplete_partitions_do_not_return_partial_rows
  on snowflake, and the corresponding six on databricks, still pass
  through unchanged paths.
• No CLI / API / UI change.

Verification

• cargo check -p utopia-server --bin utopia-server did not finish
  in the local sandbox (dev-deps cold rebuild chain on the external
  FUSE mount). The new tests are pure wiremock — no DB, no real
  cluster. Recommend running on a workstation with a normal
  filesystem before merge.

Open question for the maintainer

If the test suite already runs against test_db::url() for the
wiremock tests as well, that's a bonus; if not, direction on where
these tests should be relocated would help future PRs of the same
shape. Looking at crates/utopia-server/src/query_engine/snowflake.rs::tests
and .../databricks.rs::tests they are already inline #[cfg(test)]
and match the surrounding convention, so I'm leaving them there.

[WaylandYang]

感谢重新整理，这次只有测试本身，很清楚。

已经通过 #1221 合入 dev（c278cb8）：你的提交原样保留、署名不变，我们只补了一次 cargo fmt（lint 只差这两处换行）。你的分支没有开"允许维护者修改"，所以没法直接推到这里，这一条就先关闭了。

再次感谢。