docs: correct skyline quadrature self-consistency claim

[ivan-aksamentov]

docs/coalescent-skyline-quadrature-contract -> rust

Update:

• kb/issues/N-coalescent-skyline-quadrature-contract-undecided.md: replace the "no quadrature error" claim with self-consistency and record the off-grid error bound

• Follow-up to: Feat/exact skyline #856

The quadrature issue claimed the midpoint rule is exact for the piecewise-constant skyline, with no quadrature error. That holds only if the operative demographic function is taken to be the midpoint-snapped one. For the merger-time-midpoint boundaries the tool actually reports, the rule is self-consistent, not exact.

This corrects the wording to self-consistency and records the off-grid error characterization: the midpoint rule is exact on every lineage-count interval that contains no skyline boundary, while at most $n-1$ intervals straddle a boundary and each contributes an error bounded by its exposure times the inter-segment $T_c$ contrast [src].

Work items

• Replace the "no quadrature error" claim with self-consistency and record the off-grid error bound [src]

Possible improvements

• Implement the exact interval integral, splitting each inter-event interval at the skyline breakpoints (option O3) (kb/issues/N-coalescent-skyline-quadrature-contract-undecided.md)

[rneher]

We might want to replace this midpoint business. I'll look into it.