Implement and document custom constraints

[pnkraemer]

• Fix the display of the tutorials
• Raise errors for linearisations that don't support high-order ODEs instead of silently failing
• Introduce a probdiffeq.constraint_root_ts1 which implements custom constraints
• Write a custom_information_operator tutorial that explains how/why the custom root is beneficial
• Make dense the default SSM factorisation because it offers the most methods
• Always implement an update during initialisation to accomodate non-perfect initial Taylor coefficients (eg those for custom constraints)
• Optionally shift the STD of the initial taylor coefficients by a small value to enable updates without nans during initialisation
• Implement different ways of error normalisation (scale then norm, norm then scale) to enable constraints whose dimension doesn't match that of the ODE

Should be fully backwards compatible.