Class RenderPipeline

RenderPipeline is the one entry point that turns query SQL into the SQL that runs, with enforced ordering and a diagnostic trace. Every path uses it: saved queries, ad hoc SQL, TestQuerySQL, specs and data contexts.

Pipeline order (a data dependency, not a convention):

1. Comments are removed first, so template tags in them are inert.
2. Interpreter gate: a template must pass the MJ Query language check (QueryTemplate.Validate, Refresh mode); a template with errors is refused.
3. Render, then compose: Nunjucks renders the outer template once. Each composition it reaches records an instance (the composed query + its arguments evaluated in the caller's scope) and writes the instance's CTE name. Each instance then renders on its own, with only its arguments, recursively. The plain SQL is assembled from the rendered parts and is never rendered again, so a value that looks like a template tag stays a value.
4. Safety: the assembled statement must be a single read query (token-based; see SafetyGate).
5. Row cap / paging detect ORDER BY and inject TOP / OFFSET-FETCH / LIMIT, after rendering, because {% if SortField %} ORDER BY … {% endif %} decides whether there is an ORDER BY.