[hoegertn]

With typedErrors: true, an untyped error thrown by a resolver handler was
silently swallowed. createAppSyncHandler rethrows untyped errors, so the
Lambda invocation fails and AppSync puts the failure into the pipeline
function's ctx.error. The function's response handler returned
ctx.result without raising it, so AppSync dropped the error and the field
resolved to null with no GraphQL error and HTTP 200. Clients could not
tell a failed mutation from an empty result.

The resolver-level if (ctx.error) util.appendError(...) branch could not
catch this: an error left in a pipeline function's ctx.error never
reaches the resolver's ctx.error. Typed errors (the __appSyncError
envelope) were not affected.

Changes:

• The pipeline function now calls util.error(ctx.error.message, ctx.error.type),
  which reports the same message and Lambda:Unhandled type a UNIT
  direct-Lambda resolver does.
• Remove the dead ctx.error branch from the resolver response handler.
• Correct the typedErrors prop doc and llm.md: without the flag the
  Lambda gets no APPSYNC_TYPED_ERRORS, so typed errors rethrow and surface
  as Lambda:Unhandled, not as envelope field data.
• Add tests that execute the synthesized inline handlers with a fake
  util, covering invoke errors, typed envelopes and pass-through results.
  The invoke-error test fails against the previous handler.

Verified the new function handler with AppSync EvaluateCode on the
APPSYNC_JS 1.0.0 runtime: an unhandled Lambda error is raised, a
successful result passes through unchanged.

Deploying this changes the inline code of every typedErrors pipeline
function; the resolver topology and logical IDs are unchanged.