Fixes#4688
Instead of using generic library, build some handcrafted code to
reuse buffers, do partial parsing of the data we need for the processes
API.
Benchmark (it runs with significant number of processes on the host):
```
name time/op
PrometheusProcfs-16 3.42ms ± 8%
Processes-16 2.36ms ± 5%
name alloc/op
PrometheusProcfs-16 366kB ± 0%
Processes-16 255kB ± 0%
name allocs/op
PrometheusProcfs-16 6.76k ± 0%
Processes-16 3.83k ± 0%
```
Signed-off-by: Andrey Smirnov <andrey.smirnov@talos-systems.com>
