feat(xfa): expose raw scripts via FormSchema.Scripts; drop heuristic Rule API

[lfstokols]

Motivation

The previous XFA script-handling surface had two layers:

1. A working classification envelope (initial commit): each XFA <event> became a Rule with Type picked from the event activity, Source = field name, and a single Action{Type: ActionTypeExecute, Script: body} carrying the raw script. This part was fine — it didn't claim to interpret semantics.
2. An aspirational structured layer (Condition, Operator, LogicOp, typed Actions beyond Execute). Declared but unused until c2c3fe0 (v1.2.0), when regex-based extraction started populating it for visibility / set-value / validate / calculate patterns.

The v1.2.0 heuristic was lossy and incorrect on non-trivial scripts. Faithful FormCalc / ES5 interpretation needs a real AST, which is out of scope for a PDF library. This PR replaces the envelope with a richer, lossless one and removes the aspirational layer entirely — callers that need to evaluate scripts should plug in their own interpreter.

What changed

New surface:

• types.FormScript — verbatim script body plus event activity, language (javascript | formcalc), SOM owner path, owner ID, Name, RunAt, and a Properties map for unknown <event> / <script> attributes (listen, ref, id, binding, stateless, url, …).
• FormSchema.Scripts []FormScript — flat declaration-order list.
• Question.Scripts []string and FormSection.Scripts []string — FormScript.ID references for back-lookup.

Removed (breaking, but the only user-facing populator was the v1.2.0 heuristic):

• types.Rule, RuleType, Condition, Operator, LogicOp, Action, ActionType and all their constants.
• pdfer.Rule root-level alias (replaced by pdfer.FormScript).
• FormSchema.Rules (replaced by FormSchema.Scripts).
• Annotation.Actions (was unused).
• All XFA script-classification helpers: parseXFAScript, detectScriptLanguage, tryParseVisibilityScript, tryParseSetValueScript, tryParseValidationScript, tryParseCalculateScript, extractSimpleCondition, invertCondition, extractAllPresenceTargets, and supporting regex utilities.

<variables><script> blocks are now exposed verbatim with Event="variables" rather than being chopped into per-function pseudo-rules. Language defaults to formcalc per the XFA spec when contentType is absent.

Name overlap: forms/acroform/actions.go still defines its own ActionType for PDF action dictionaries (URI, JavaScript, …). That type is unrelated to the deleted XFA ActionType and stays.

Migration

For callers that used the old Rule/Action surface — the field correspondence is:

Old	New
Rule.ID (rule_N)	FormScript.ID (stable: SOM path + event + index)
Rule.Type (bucket: calculate/validate/setValue)	FormScript.Event (literal XFA activity)
Rule.Source	FormScript.OwnerID / FormScript.OwnerPath
Action.Script	FormScript.Body
none	FormScript.Language, Name, RunAt, Properties

Typical usage:

for _, s := range schema.Scripts {
    fmt.Printf("[%s] %s on %s\n%s\n", s.Event, s.Language, s.OwnerPath, s.Body)
}

Known gaps

Scripts attached to nodes pdfer does not surface in the schema are not extracted: decorative <draw> elements with events, <field> buttons with bind="none" other than AddAttachment, <pageArea>-level events, and individual <field> radio options collapsed into an <exclGroup>'s Options. Documented on the FormScript type comment. Callers needing full event fidelity should walk the raw XFA XML directly.

Verification

• go test ./... — passes.
• End-to-end on tests/resources/nonivd_estar.pdf: 769 scripts emitted, bodies preserved verbatim (including comments, escapes, multi-line if/else chains), 621/904 questions carry script ID references.

Test plan

• go test ./...
• XFA package tests cover envelope (event/script Properties capture, <variables><script> branch, Question/Section script ID resolution, language defaulting, name/runAt plumbing)
• eSTAR roundtrip script counts (769 / 887 / 649) unchanged from pre-ripout

🤖 Generated with Claude Code