fix(describegpt): honor QSV_LLM_BASE_URL env var; repair stale describegpt tests

[jqnatividad]

Fixes 5 describegpt integration tests that fail under QSV_TEST_DESCRIBEGPT=1 against a live local LLM. Surfaced while running the test suite for #3888; unrelated to that PR, so split out here.

Real bug — QSV_LLM_BASE_URL was silently ignored

The DEFAULT_BASE_URL constant (https://api.openai.com/v1) had drifted from the actual --base-url docopt default (http://localhost:1234/v1). Two base-URL precedence checks compare flag_base_url against that sentinel to tell "user explicitly passed --base-url" from "docopt default in effect". Because the sentinel never matched the real default, the comparison was always true — describegpt always took the "explicit flag" branch, used the docopt default, and never consulted QSV_LLM_BASE_URL. The env-var precedence branch was dead code.

Fix: realign DEFAULT_BASE_URL (and the default prompt file's base_url) with the docopt default, so the intended explicit --base-url > QSV_LLM_BASE_URL > default precedence works. Added a comment tying the constant to the docopt default so it can't drift again.

Fixes describegpt_baseurl_precedence_env_over_default.

Test repairs

• describegpt_stats_options_file_prefix / describegpt_both_file_prefixes — the hand-written stats.csv fixture had an empty nullcount cell in both data rows; parse_stats_csv parses nullcount strictly and errored. Filled in the missing 0s.
• describegpt_empty_dataset — expected success on a header-only CSV, but describegpt's first step (stats) cannot compile statistics for a zero-row file. Changed it to assert the (correct) error.
• describegpt_score_high_threshold_triggers_retries — accepted only 2 of the retry-path messages; broadened to all of them, since which message appears depends on the non-deterministic refined query.

Verification

Run against a live local LLM (LM Studio, QSV_TEST_DESCRIBEGPT=1):

• All 5 previously-failing tests now pass.
• Full describegpt integration suite: 70 passed, no regressions. (describegpt_jsonschema_dictionary fails intermittently under sustained serial load — pre-existing LLM-output flakiness, passes on re-run.)
• 81 describegpt unit tests pass.

🤖 Generated with Claude Code

[codacy-production]

Up to standards ✅

🟢 Issues 0 issues

Results:
0 new issues

View in Codacy

🟢 Metrics 7 complexity

Metric	Results
Complexity	7

View in Codacy

_{NEW Get contextual insights on your PRs based on Codacy's metrics, along with PR and Jira context, without leaving GitHub. Enable AI reviewer
TIP This summary will be updated as you push new changes.}