wildedge-lint: custom lint rules for misuse detection

[piotrekno1]

Add a wildedge-lint module bundled inside the existing AAR. Consumers get IDE warnings with no extra dependency.

Checks

Tier 1 (method chain, low false-positive risk)

• Task<T> terminal call without .trackWith() in the chain
• generateContentStream() without .trackWith() in the chain
• trackInference() not inside a trace {} lambda

Tier 2 (data flow, defer until Tier 1 ships)

• Interpreter, OrtSession, InterpreterApi instantiation where the result is never passed to wildEdge.decorate()

Tier 2 requires tracking a variable through field storage and cross-method use. Naïve detection produces false positives on correctly-decorated code. Track in a follow-up once there is a clean strategy.

Acceptance criteria

• Tier 1 checks as Detector + IssueRegistry
• Checks fire in Android Studio and ./gradlew lint
• Zero false positives on the sample app
• Tier 2 tracked in a follow-up issue