Clarify #333 performance impact

[ValarDragon]

Summary

• update Use dedicated squaring for public powers and fixed-base windows #333's repeated-squaring figures with measurements from the narrowed
  47cf57d candidate, covering both Fp and Fq;
• describe generic fixed-base preprocessing by its exact operation-count
  reduction instead of retaining measurements from the superseded broader
  draft;
• clarify that Orchard uses precomputed coefficient tables and therefore does
  not receive a key-generation speedup from that generic table generator.

At k = 11, dedicated repeated squaring measured 64.1–64.4x faster on Apple
M4 and 64.9–65.0x faster on x86_64 Linux. For an 85-window table, the
fixed-base change replaces 764 constant-time exponentiations with 168 scalar
multiplications.

Validation

• ./scripts/changelog.py check
• git diff --check
• changelog lines are at most 80 columns

This is documentation-only and makes no API changes.

[v12-auditor]

Warning

No auditable source files found in this PR's diff.

[ValarDragon]

Closing because the changelog policy correctly prevents one PR from modifying another PR’s fragment. The focused measurements and exact fixed-base operation count are now recorded in merged PR #333’s description; no code correction is required.