Bump for correlation

[MarcusRainbow]

There are two basic types of bump we'd like to support:

1. A bump to all correlations, designed to preserve positive-semi-definiteness of the correlation matrix.
2. A bump to one correlation pair. This is quite likely to pull the matrix into non-positive-semi-definite territory, so we'd need some way to handle that, for example bumping the pair in the other direction or by a different amount, which we'd have to report.