[MISC] Extend insert_iterator

[eseiler]

For our compute_sketches benchmark, this is around 10-14 times faster.

We could also add an overload for the ibf, which then just calls emplace.
This could be used in insert_into_ibf and the IBF ctor, where we currently first put the hashes in a set or vector before inserting.

Numbers are iterations.
Clang apparently optimizes way better than GCC.
Clang with inline = remove [[gnu::noinline]] from void invoke_without_inlining

Benchmark	GCC before	GCC	Clang before	Clang	Clang with inline
manual_construct/64	32	32	44	54	41
manual_construct/128	34	34	52	54	42
manual_construct/256	34	34	54	54	44
manual_construct/512	33	33	51	54	44
manual_construct/1024	34	28	52	54	44
config_construct/64	8	21	10	28	14
config_construct/128	9	20	11	30	16
config_construct/256	9	20	11	30	16
config_construct/512	9	18	11	29	16
config_construct/1024	9	16	12	26	15
config_and_max_construct/64	13	27	17	39	23
config_and_max_construct/128	13	28	17	39	23
config_and_max_construct/256	13	28	18	40	23
config_and_max_construct/512	11	28	17	40	24
config_and_max_construct/1024	13	27	18	40	24

[seqan-actions]

Documentation preview available at https://docs.seqan.de/preview/seqan/hibf/237

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.58%. Comparing base (6e1fd5a) to head (76202c1).
Report is 5 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #237      +/-   ##
==========================================
- Coverage   99.63%   99.58%   -0.05%     
==========================================
  Files          51       51              
  Lines        1929     1951      +22     
  Branches        5        5              
==========================================
+ Hits         1922     1943      +21     
- Misses          7        8       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[smehringer]

ah sorry! one more thing: When moving to a fast layout or partitioned IBF I need to compute not only hyperloglog sketches but at the same time minHash sketches. Before I had the kmer set and then computed the respective sketches from that. Now I could use the function type I guess? Or I could still use the set.

(min hash sketches is a small heap that gets updated with every new kmer for the XXX smallest kmers in the set).

[smehringer]

How can this function be constexpr? The type in the switch case is not?

[eseiler]

The type would be known in a constant expression, but what isn't allowed in constexpr is casting from void *.
Made it inline instead.

[eseiler]

ah sorry! one more thing: When moving to a fast layout or partitioned IBF I need to compute not only hyperloglog sketches but at the same time minHash sketches. Before I had the kmer set and then computed the respective sketches from that. Now I could use the function type I guess? Or I could still use the set.

(min hash sketches is a small heap that gets updated with every new kmer for the XXX smallest kmers in the set).

Didn't really get faster when using a function.
The main thing is that you cannot get rid of kmers, because you still need it.