chore: PRAGMA synchronous is unimplemented, silently ignored

[iheitlager]

Description

src/vdbe/pragma.rs has no handling for PRAGMA synchronous — it's silently ignored rather than rejected or acted upon, so every commit unconditionally does full fsyncs (matching only SQLite's synchronous=FULL default) and there is no way to opt into NORMAL/OFF semantics. This is a parity gap against real SQLite, not the cause of the current benchmark's 10x gap (both benchmark sides use library defaults, so it's apples-to-apples) — filed separately as lower priority.

Complexity

Estimate: medium
Reasoning: Requires deciding how synchronous interacts with the existing rollback-journal/WAL fsync call sites (src/pager.rs:589,597,697) — not just a parse-and-store no-op, since it should actually change fsync behavior for NORMAL/OFF to be meaningful, which touches pager commit logic and needs spec/ADR consideration for durability semantics.

Context

Surfaced during the same 4-agent investigation as the DDL-reparse and tokenizer tickets, but is an independent parity gap rather than a perf-bug contributor.

Acceptance Criteria

• PRAGMA synchronous is recognized and stored per-connection
• FULL/NORMAL/OFF map to the correct fsync call pattern in the pager
• Oracle-parity test confirms PRAGMA synchronous round-trips (PRAGMA synchronous query returns the set value)

[iheitlager]

Closed via #650.

Implemented FULL/NORMAL/OFF with the fsync-skip policy documented in ADR-0036. Two notable implementation decisions:

• SynchronousMode lives in src/header.rs, not src/pager.rs (its otherwise natural home) — vdbe/pragma.rs needs to name it directly, and vdbe/ may never use crate::pager directly (spec 001-architecture Requirement 1, enforced by tests/unit/layer_isolation.rs). Same reason JournalMode already lives there.
• Along the way, found and fixed that repl.rs's generic (non-SELECT) statement path was silently discarding any result rows a program emitted — meaning PRAGMA integrity_check's output was already being dropped. Pre-existing and unrelated to this ticket, but fixed alongside it since synchronous's query form needed the same row-printing to be testable via the CLI at all.

Oracle-parity round trip test added at tests/unit/pragma_synchronous_repl.rs, diffing our repl output against real sqlite3 for PRAGMA synchronous = <level>; PRAGMA synchronous;.

spend: matched estimate (medium)

---

🤖 Analysis by Claude