fix: align pull_request_html input name across pr-ask-template.yml

The input was declared as pull_request_html_url but referenced as
inputs.pull_request_html internally, and both caller workflows
(pr-ask-ive.yml, pr-ask-reply-ive.yml) pass it as pull_request_html -
none of the three names matched. PULL_REQUEST_HTML_URL was always empty
inside the container as a result, so every /ask call failed on an empty
--pr-url before ever reaching the LLM. Likely introduced by 9ca66c2
(Fixing the URL names), which renamed this input on pr-review-template.yml
but only half-applied the same rename here.

.gitea/workflows/pr-ask-template.yml

@@ -18,7 +18,7 @@ on:
         required: true
         type: "string"
 
-      pull_request_html_url:
+      pull_request_html:
         description: "html url of the pull request being asked about"
         required: true
         type: "string"
@@ -35,6 +35,15 @@ jobs:
           ISSUE_NUMBER: ${{ inputs.issue_number }}
           COMMENT_ID: ${{ inputs.comment_id }}
           PULL_REQUEST_HTML_URL: ${{ inputs.pull_request_html }}
+          # (env var name kept as PULL_REQUEST_HTML_URL for clarity in the
+          # run: script below - this maps from the `pull_request_html`
+          # input, which is what both pr-ask-ive.yml and
+          # pr-ask-reply-ive.yml actually pass via `with:`. A prior version
+          # of this file declared the input as `pull_request_html_url` but
+          # referenced `inputs.pull_request_html` here - neither name
+          # matched the caller's `with:` key, so PULL_REQUEST_HTML_URL was
+          # always empty and --pr-url="" made every /ask call fail before
+          # ever reaching the LLM.
         run: |
           question="${COMMENT_BODY#/ask}"
           question="$(echo "$question" | sed -e 's/^[[:space:]]*//')"