Raise PoWAveragingWindow to 51 at NU7 activation

[evan-forbes]

Summary

Adds the difficulty-averaging-window change to the blocktime-reduction ZIP. At NU7 activation, PoWAveragingWindow switches from 17 to 51, preserving the 1,275-second wall-clock smoothing window when target spacing drops from 75 s to 25 s.

• Consensus change: introduces PostNU7PoWAveragingWindow := 51 and makes PoWAveragingWindow a height-dependent function (17 pre-NU7, 51 post-NU7). Mirrors the height-dependent pattern already used for PoWTargetSpacing.
• Effect on difficulty adjustment: updated inline to describe the change ($17 \times 75 = 1{,}275$ s → $17 \times 25 = 425$ s contraction without the bump, vs. $51 \times 25 = 1{,}275$ s preserved with it).
• Rationale — Difficulty averaging window: new section with a Scope note (51 is chosen to preserve the status quo, not to claim it is the optimal value of PoWAveragingWindow), Zebra-simulator recovery thresholds, the Blossom $2{,}550 \to 1{,}275$ s precedent, and a devnet stddev measurement.
• Footnote: points at benchmark_hash_rate_shock_daa_configurations and simulate_three_x_target_spacing_reduction in the Zebra benchmark branch.

Draft for internal review on the valargroup fork before opening upstream against zcash/zips:main.

Test plan

• make all-zips via the CI render docker image succeeds (exit 0).
• In-document cross-refs [Difficulty averaging window](#difficulty-averaging-window) and [Averaging window](#averaging-window) resolve to the rendered anchor IDs (explicit MMD labels on those headings).
• Recovery table cells with ~ render as literal tildes (escaped to \~ so MMD does not interpret them as <sub> markers).
• Footnote [^daa-recovery-benchmark] renders with both function-level GitHub line anchors.
• Reviewer to confirm wording / scope framing reads correctly.