fix(transpiler): trailing var returns, receiver history reads, switch na fallback

[yusufcansehirli]

Three codegen fixes (same batch as #318)

All three reproduced with public Library studies running through the engine.

1. A trailing var declaration is not returned

f() => var fibs = array.from(...) returns undefined: the declaration is emitted but no implicit return is generated, so callers doing f().get(i) crash with Cannot read properties of undefined (reading 'get').
Fix: keep the declaration (hoisting/series registration) and append an implicit return <name>.

2. History reads in receiver chains stay raw

(arr[2]).get(i) inside an operand emitted $$.let.arr[2].get(i) — a raw bracket on the Series wrapper object (undefined at runtime). The same expression in a ternary consequent lowered correctly to $.get($$.let.arr, 2).get(i).
Fix: resolveCalleeObject now lowers computed receivers to $.get(obj, offset); transformOperand does the same for bare computed member operands.

3. Unmatched switch expression returns undefined

A discriminated switch expression with no matching arm (or no default case) resolved to undefined, crashing tuple destructuring on [0]. Pine semantics: na.
Fix: both switch-IIFE generators append a return na; fallback when there is no default arm, and tuple-destructuring element reads use optional chaining ($.get(temp, 0)?.[n]) so an na value resolves instead of crashing.

Verification

• Targeted probes: trailing-var returns the array (size() == 3); unmatched-switch destructure runs; operand history read yields a finite number (no NaN).
• tests/transpiler/ — 504 tests pass (one snapshot updated for the ?.[n] reads).
• 13 previously-failing Library sources swept through the real engine: 11/13 OK (2 remaining failures are source-side artifacts unrelated to the engine).
• Isolated re-runs of the touched areas are clean; intermittent streaming/timing test flakes reproduce on unpatched main in this environment.