docs(adr): durability, codec, metrics, tenet gates, encoding 0014-0018 (closes #59, #53, #7, #157, #37)

[ELares]

The final foundational decisions.

What

• 0014 ([DECISION]: Durability stance (ephemeral default, opt-in snapshot, warm-restart, later tiers) #59) ephemeral default; opt-in forkless snapshot + mmap warm-restart; SSD/append-log later (Garnet-style safe defaults).
• 0015 ([DECISION]: Default codec = zstd low-level; LZ4 and none as policy options #53) zstd-low default codec; LZ4/none per-keyspace (the transparent compression Redis refuses to do).
• 0016 ([DECISION]: Headline metrics are throughput-per-core and memory-at-fixed-hit-ratio #7) per-core throughput + memory-at-fixed-hit-ratio + p99/p999, open-loop/coordinated-omission-corrected; rejects aggregate QPS.
• 0017 ([DECISION]: Per-tenet acceptance targets and release gates for Compatible/Efficient/Simple/Scalable/AI-Driven #157) a measurable release gate per tenet (Compatible/Efficient/Simple/Scalable/AI-Driven).
• 0018 ([RESEARCH]: Adaptive vs fixed encoding-conversion thresholds #37) fixed Redis-compatible encoding thresholds; adaptive deferred to the advisor (resolves the path-bug-blocked research with a grounded decision).

Review focus

• Each Decision matches its issue; ADR-0016 methodology is honest (per-core, coordinated-omission); ADR-0017 gates are consistent with ADR-0009/0012/0013/0016.
• [RESEARCH]: Adaptive vs fixed encoding-conversion thresholds #37 closure: deciding it (vs leaving the blocked research open) is justified because the thresholds are grounded in pinned claims, not the missing doc.
• Citations exist + match; INDEX lists 0014-0018; no em/en dashes.

Closes #59. Closes #53. Closes #7. Closes #157. Closes #37. Part of M0: Vision and Scope.