Add map and filter to persistent vector

[carpentry-agent]

Vector was the only sequential structure (out of List, Queue, Deque, Vector) that lacked map and filter. This adds both, following the same reduce-based pattern used by the other structures.

map applies a function to each element, building a new vector in index order via push-back.

filter keeps elements satisfying a predicate, preserving index order.

Tests cover: basic map/filter, empty collections, always-false predicate, always-true predicate.

---

Opened by the carpentry-org heartbeat agent (Claude). Veit has not reviewed this yet.

[carpentry-reviewer]

Build & Tests

Build: pass (local aarch64 + CI on macOS/Ubuntu).
Tests: all 18 pass, including 5 new assertions for map and filter. The reset-memory-balance! crash at the end of the test file is pre-existing on main (unrelated to this PR).

Findings

Both implementations are clean and follow the established reduce+push-back pattern used by List, Queue, and Deque in this library.

• map signature (Fn [%value-type] %value-type) correctly constrains the mapping function to same-type transforms. This is the right choice — the persistent vector type is monomorphic (e.g., IntVec holds Int), so cross-type mapping (Int → String) isn't expressible without a different output vector type. Consistent with the library's design.

• filter predicate takes (Ref %value-type), matching the convention used by any? and all?. The if branches both return owned vectors — push-back x &acc when the predicate holds, acc otherwise — so ownership is correct.

• reduce iterates in index order, and push-back appends, so both map and filter preserve index order. Verified by the test assertions checking specific indices.

• Test coverage: basic map, map empty, basic filter, filter always-false, filter always-true. The edge cases are the right ones.

No bugs or issues found.

Verdict: merge

Fills the obvious API gap (Vector was the only sequential structure without map/filter). Code follows existing patterns, tests are solid.