Optimize reading of old style values

PR #769 introduced the reading of old style values. Currently all the data is read into old `Values` and then converted into `Variants`. Additionally to read the uncertainty all Values are read and then only the uncertainty from the first used.

Start by reviewing PR #769 and the current old-style value-reading path. Trace how Values are read and converted into Variants, including uncertainty handling; done means avoiding unnecessary full reads while preserving the required values and uncertainty behavior.

Written by the indexing model from the issue text.