Per-cpu for histogram metrics

@bobrik, hi!

I'm trying to monitor per-CPU latency histograms for NET_RX softirqs using a configuration inspired by the percpu-softirq example. The idea is to observe histogram distributions (e.g., softirq_wait_seconds and softirq_runtime_seconds) on a per-CPU basis in environments where different CPUs handle traffic from different network interfaces.

I used the following config snippet:

```
metrics:
histograms:
- name: softirq_wait_seconds
help: Latency histogram for how long it takes from NET_RX softirq raise to entry aka wait time
bucket_type: exp2
bucket_min: 0
bucket_max: 31
bucket_multiplier: 0.000000001 # nano to seconds
labels:
- name: bucket
size: 4
decoders:
- name: uint
- name: cpu
size: 2
decoders:
- name: uint
```
However, the output only contains the bucket label, and not cpu, like this:

```
ebpf_exporter_softirq_runtime_seconds_bucket{bucket="6",le="0.536870912"} 0
ebpf_exporter_softirq_runtime_seconds_bucket{bucket="6",le="1.073741824"} 0
ebpf_exporter_softirq_runtime_seconds_bucket{bucket="6",le="2.147483648"} 0
ebpf_exporter_softirq_runtime_seconds_bucket{bucket="6",le="+Inf"} 0
```
Expected behavior would be to have:

```
ebpf_exporter_softirq_runtime_seconds_bucket{cpu="3",le="0.536870912"} 0
...
```
Question:

Does ebpf_exporter currently support per-CPU labels for histograms (similar to how it works for counter types in the percpu-softirq example)? If not, is it something that could be added, or is there a suggested workaround?

Thanks in advance for any guidance!

Start by comparing the histogram configuration with the percpu-softirq example and trace how histogram labels are handled. Confirm whether per-CPU histogram labels are supported; done means exposing the cpu label in histogram output, with tests covering the expected metric labels.

Written by the indexing model from the issue text.