Optimized metric name extraction in distributor

[pracucci]

What this PR does:
#3990 had the subtle con that the series labels are retained in the ValidationError. Instead of trying to solve it (which could undo the performance benefit got in that PR) I'm proposing to document it and leverage on it to further optimize the metric name extraction.

name                                                      old time/op    new time/op    delta
Distributor_Push/too_many_labels_limit_reached-12           1.26ms ± 1%    1.23ms ± 0%   -2.13%  (p=0.008 n=5+5)
Distributor_Push/max_label_name_length_limit_reached-12     4.97ms ± 1%    4.95ms ± 0%     ~     (p=0.190 n=5+4)
Distributor_Push/max_label_value_length_limit_reached-12    2.96ms ± 0%    2.94ms ± 2%     ~     (p=0.190 n=4+5)
Distributor_Push/timestamp_too_old-12                       1.02ms ± 0%    0.97ms ± 0%   -5.21%  (p=0.008 n=5+5)
Distributor_Push/timestamp_too_new-12                       1.02ms ± 1%    0.97ms ± 2%   -4.87%  (p=0.008 n=5+5)
Distributor_Push/all_samples_successfully_pushed-12         1.18ms ± 1%    1.13ms ± 1%   -4.37%  (p=0.008 n=5+5)
Distributor_Push/ingestion_rate_limit_reached-12             874µs ± 0%     821µs ± 0%   -6.08%  (p=0.008 n=5+5)

name                                                      old alloc/op   new alloc/op   delta
Distributor_Push/too_many_labels_limit_reached-12           94.4kB ± 0%    91.2kB ± 0%   -3.37%  (p=0.008 n=5+5)
Distributor_Push/max_label_name_length_limit_reached-12      129kB ± 0%     126kB ± 0%   -2.53%  (p=0.008 n=5+5)
Distributor_Push/max_label_value_length_limit_reached-12     130kB ± 0%     126kB ± 0%   -2.51%  (p=0.008 n=5+5)
Distributor_Push/timestamp_too_old-12                        118kB ± 0%     111kB ± 0%   -5.59%  (p=0.008 n=5+5)
Distributor_Push/timestamp_too_new-12                        118kB ± 0%     111kB ± 0%   -5.64%  (p=0.008 n=5+5)
Distributor_Push/all_samples_successfully_pushed-12          168kB ± 0%     162kB ± 0%   -3.75%  (p=0.008 n=5+5)
Distributor_Push/ingestion_rate_limit_reached-12            85.2kB ± 0%    78.8kB ± 0%   -7.50%  (p=0.008 n=5+5)

name                                                      old allocs/op  new allocs/op  delta
Distributor_Push/too_many_labels_limit_reached-12            3.13k ± 0%     2.13k ± 0%  -31.93%  (p=0.000 n=5+4)
Distributor_Push/max_label_name_length_limit_reached-12      3.08k ± 0%     2.08k ± 0%     ~     (p=0.079 n=4+5)
Distributor_Push/max_label_value_length_limit_reached-12     3.08k ± 0%     2.08k ± 0%  -32.47%  (p=0.008 n=5+5)
Distributor_Push/timestamp_too_old-12                        5.03k ± 0%     3.03k ± 0%  -39.75%  (p=0.008 n=5+5)
Distributor_Push/timestamp_too_new-12                        5.03k ± 0%     3.03k ± 0%  -39.75%  (p=0.008 n=5+5)
Distributor_Push/all_samples_successfully_pushed-12          4.04k ± 0%     2.04k ± 0%  -49.55%  (p=0.000 n=5+4)
Distributor_Push/ingestion_rate_limit_reached-12             4.03k ± 0%     2.03k ± 0%  -49.62%  (p=0.008 n=5+5)

Which issue(s) this PR fixes:
N/A

Checklist

• Tests updated
• Documentation added
• CHANGELOG.md updated - the order of entries should be [CHANGE], [FEATURE], [ENHANCEMENT], [BUGFIX]

[pstibrany]

Unstoppable! Is there more performance improvements to squeeze out?

[pracucci]

Is there more performance improvements to squeeze out?

There's always room for improvement, but I will stop here for now :)