docs: correct df.instance_nodes column types to TEXT

[pinodeca]

Problem

df.instance_nodes() documents result and status_details as JSONB, but
the function actually returns them as TEXT.

The values originate from genuine JSONB columns on df.nodes, but the
function explicitly casts them on the way out:

• result::text in the SPI projection (src/monitoring.rs)
• "status_details::text" returned by status_details_select_expr
  (src/node_status.rs)

This also made the documented example non-functional. ->> is only defined
for json/jsonb, so copy-pasting the documented query:

SELECT status_details->>'execution_id' FROM df.instance_nodes('a1b2c3d4');

fails with operator does not exist: text ->> unknown.

Changes

• docs/api-reference.md: column types for result and status_details
  corrected from JSONB to TEXT, with a note that they carry JSON text and
  need an explicit ::jsonb cast for JSON operators.
• docs/api-reference.md: the example query now casts before extracting.
• USER_GUIDE.md: the Function Nodes column descriptions note the same cast
  requirement.

Documentation only — no code, schema, or upgrade-script changes.

Follow-up (not in this PR)

Whether TEXT is the right return type is a separate question. The sibling
APIs (df.result(), df.instance_info().output) return TEXT because their
payload comes back from the duroxide runtime as an opaque string with no schema
guarantee. That reasoning does not apply here: these two columns come from real
JSONB columns whose contents are guaranteed well-formed, so the cast is pure
ceremony and converts an impossible failure into a possible one.

Changing it is a breaking signature change (DROP FUNCTION + CREATE, plus a
retained TEXT-returning wrapper symbol for Scenario B1 against frozen 0.2.2-0.2.4
schemas). Worth deciding while the 0.2.5 upgrade script is still open, since the
API is only one release old and the pre-1.0 breaking-change allowance goes away
at 1.0.

Closes #295