Fixes #3761

parsePoolObjsetFile finds the dataset name by searching the whole kstat line for the third whitespace-separated field:

datasetName = line[strings.Index(line, parts[2]):]

When a root dataset's name also occurs inside the dataset_name field name, strings.Index matches that prefix instead of the value column and the label ends up holding the entire input line. For a pool called data:

node_zfs_zpool_dataset_nwritten{dataset="dataset_name                    7    data",zpool="data"} 0

Child datasets were unaffected because e.g. data/vm-1101-disk-0 doesn't occur in the prefix, which is why this only shows up on the pool root.

The fix walks past the field name and type columns first, then takes the remainder. Those columns come from strings.Fields, so they're non-whitespace tokens and can't match inside the leading separator — the value offset is unambiguous.

I deliberately didn't use strings.Join(parts[2:], " ") as suggested in the issue. Fields collapses runs of whitespace, so re-joining with a single space would rewrite the existing pool3/dataset with space fixture to one space and silently corrupt label values for datasets that really do contain consecutive spaces. Keeping the remainder verbatim preserves them byte-for-byte.

Added a data pool fixture reproducing the collision (root plus a child dataset), extended TestZpoolObjsetParsingWithSpace, and regenerated both e2e outputs.

Fixes #3761