AI Expression malformed individual experiment responses cause user-facing exception

Every now and then, the user will get a
We're sorry, something went wrong.
during the processing of a gene.

I managed to track one down and it's malformed JSON returned by OpenAI despite it having been given a responseFormat schema. Boo!

I think the only solution is to assume these are 1 in 50 events and have a retry strategy?

From the tomcat logs for VectorBase on watermelon

Caused by: java.lang.RuntimeException: Error parsing JSON response for dataset DS_047c56768c.  Raw response string:
{"one_sentence_summary":"This gene is moderately expressed in <i>Anopheles gambiae</i> midguts and salivary glands infected with <i>Plasmodium falciparum</i>, showing a slightly h
igher expression in salivary glands.","biological_importance":3,"confidence":4,"notes":"Expression values indicate a slightly higher level in salivary gland samples compared to mi
dgut samples, and expression seems consistent across paired-end and single-end methodologies.","experiment_keywords":["mosquito tissues","midgut","salivary gland","Plasmodium falc
iparum infection","RNA-Seq"],"confidence":4}

        at org.apidb.apicommon.model.report.ai.expression.Summarizer.lambda$describeExperiment$0(Summarizer.java:159)

Caused by: org.json.JSONException: Duplicate key "confidence" at 579 [character 580 line 1]
        at org.json.JSONTokener.syntaxError(JSONTokener.java:536)
        at org.json.JSONObject.<init>(JSONObject.java:259)
        at org.json.JSONObject.<init>(JSONObject.java:481)
        at org.json.JSONObject.<init>(JSONObject.java:458)
        at org.apidb.apicommon.model.report.ai.expression.Summarizer.lambda$describeExperiment$0(Summarizer.java:150)

Parsing is here: https://github.com/VEuPathDB/ApiCommonWebsite/blob/8c87e63cef7f9e691d377fe80bc7177f70cba6bc/Model/src/main/java/org/apidb/apicommon/model/report/ai/expression/Summarizer.java#L150

See also #263

Start at Model/src/main/java/org/apidb/apicommon/model/report/ai/expression/Summarizer.java around line 150, where the experiment response is parsed, and compare the related context in issue #263. Reproduce the duplicate-key malformed response if possible and determine how parsing should recover or retry. Done means malformed individual responses no longer surface as an unhandled user-facing exception.