acedrg_tables: extract insert_bond_row / insert_angle_row

[pemsley]

Extract the inline per-row index-insertion logic from load_bond_tables() and load_angle_tables() into two public methods, insert_bond_row() and insert_angle_row(), and point the loaders at them.

This deduplicates the key construction and gives external code a way to populate the bond/angle index containers row by row (e.g. an on-demand backend feeding fill_restraints without bulk-loading the whole table set). The scratch key buffers are passed in by the caller and reused across rows, avoiding per-row allocations that the old inline code made.

Behaviour-preserving: gemmi drg output is byte-identical before and after - tested against a handful of monomers.

[keitaroyam]

I see key_buf and hybr_buf are used for performance reasons. Have you measured how much faster this is compared to defining them locally inside the function? I ask because it forces the caller to manage temporary variables.

If defining them inside the function turns out to be significantly slower, could we move key_buf and hybr_buf to be private member variables of AcedrgTables instead?

[pemsley]

Thank you for your feedback - I will get back to you.

[pemsley]

Yes, for performance reasons.

Trial-No:   -pre-bufs      +pre-bufs   delta
 1           10.551           9.620    0.931
 2           10.260           9.504    0.756
 3           10.597           9.629    0.968
 4           10.438           9.632    0.807
 5           10.521           9.735    0.786
 6           10.464           9.769    0.695
 7           10.283           9.768    0.516
 8           10.618           9.699    0.919

My understanding is that you don't much like key_buf and hybr_buf because are an implemenentational detail not appropriate for a public signature.

I will rework the commit so that they are private members.