fix(amend): evaluate first arg; in-place amends return the symbol

[singaraiona]

Problem

insert/upsert decided in-place vs functional by inspecting the raw parse node for a literal 'sym (the ATTR_QUOTED flag), so only a literal first arg drove in-place mode. A symbol arriving any other way raised error: type:

(set t (table [a] (list [])))
(insert 't (list 1))      ;; ok — literal 'sym
(set name 't)
(insert name (list 2))    ;; error: type   ← symbol via a variable

update and alter already did the right thing (evaluate the first arg, resolve a symbol to its named global).

Fix

Unify all amends on one rule: evaluate the first argument; a symbol value names a global to amend in place, a table/vec value is amended functionally. In-place amends now return the symbol rather than the amended clone (matching v1).

• query.c — insert/upsert mirror update's eval-first dispatch; dropped the ATTR_QUOTED raw-node special case. In-place returns the symbol across insert/upsert/update. (The internal upsert→insert recursion via the already-evaluated table path is untouched.)
• tblop.c — alter in-place returns the symbol.

This works through variables, (quote t), and lambda parameters, since everything now routes through ray_env_get/ray_env_set like the literal 'sym path always did.

Tests

• New regression test/rfl/table/insert_inplace_symbol.rfl (variable/quote/lambda forms, functional-vs-in-place, symbol return value across insert/upsert/update/alter).
• Updated existing tests that relied on the old value-return behavior (update.rfl, query_coverage.rfl, linkop/coverage.rfl, test_lang.c), including the (set x (alter 'x ...)) idiom which now uses the in-place mutation directly.

Full suite passes under the sanitizer build (3441 passed, 0 failed, 3 pre-existing skips).

🤖 Generated with Claude Code