ZFS collector misparses dataset values that occur in the kstat field prefix

Host operating system

Linux 7.0.14-4-pve x86_64, OpenZFS 2.4.3

node_exporter version

Reproduced with both:

• node_exporter 1.9.1
• node_exporter 1.12.1

The same parsing code is present on the current default branch.

What did you do?

Created/read the root dataset of a pool named data. Its objset kstat contains
a valid string field:

dataset_name                    7    data

The child dataset data/vm-1101-disk-0 is parsed correctly, so this is specific
to a dataset value that occurs earlier in the complete input line.

What did you expect to see?

node_zfs_zpool_dataset_nwritten{dataset="data",zpool="data"} ...

What did you see instead?

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

Cause

parsePoolObjsetFile currently extracts the dataset with:

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

For the value data, strings.Index matches the data at the beginning of
dataset_name, rather than the value in the third column. The resulting label
is therefore the entire input line. Joining parts[2:] would retain dataset
names containing spaces without searching the field-name/type prefix.

Start with parsePoolObjsetFile, using the reported root dataset case where the value is data and also appears in dataset_name. Check the parsing and add or update coverage for the resulting node_zfs_zpool_dataset_nwritten label; done means the root dataset is labeled dataset="data" while child dataset parsing remains correct.