Replace O(n²) String.concat in RedisPipeline.add with StringBuf

[carpentry-agent]

RedisPipeline.add rebuilt the entire buffer string on every call via
String.concat, making pipeline construction O(n²) in command count.

This replaces the buf field type from String to StringBuf (from
carpentry-org/strbuf), which grows geometrically and gives amortized
O(1) appends.

Changes:

• Add strbuf@0.1.0 dependency
• Change RedisPipeline field buf from String to StringBuf
• begin now creates a StringBuf instead of an empty string
• add uses StringBuf.append-str instead of String.concat
• pipeline-exec converts the buffer to String for TcpStream.send
• Tests updated to use StringBuf.str/StringBuf.length for assertions

---

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

[carpentry-reviewer]

Build & Tests

Build: pass (local, aarch64-linux)
Tests: 77/77 pass
CI: pass (macOS + Ubuntu)

One minor warning during compilation: [Warning] No 'prn' function found with type (Fn [StringBuf] String) — cosmetic, since StringBuf doesn't derive prn. Not a blocker.

Findings

Clean, focused change. Reviewed redis.carp:544-597 and test/resp.carp diff in full.

• The StringBuf.append-str call at redis.carp:563 correctly mutates the buffer through (buf p), and set-count! updates the count. Ownership is sound.
• In pipeline-exec at redis.carp:596, StringBuf.str converts the buffer to a String for TcpStream.send. The temporary String lives for the duration of the expression, so the & reference is valid. Correct.
• Test updates properly use StringBuf.str to extract the contents and StringBuf.length for the empty-buffer check. Matches the new types.
• No callers of RedisPipeline.buf outside the test file access the raw field, so the type change is self-contained.

No bugs, edge cases, or regressions found.

Verdict: merge

Straightforward performance fix. The O(n²) → O(1) analysis is correct, the implementation is clean, and the tests cover the same scenarios as before with the new types.