dbind() of a named list

[benthestatistician]

When dbind()'s first arg is a named list of ISMs, it should used those names as the levels of @groups in the BISM it produces.

[benthestatistician]

A special bonus to this enhancement would be that when it's given a list of ISM and BISMs, it pulls levels for the non-BISM ISMs from the list names while pulling levels corresponding to the BISMs from levels of those BISMs' @groups.

[nullsatz]

I was nervous about doing this myself because I felt that I didn't really understand what a block diagonal matrix was and its relationship to the group slot. I had claude explain it to me and also implement the requested changes since I had procrastinated well past the "best if used by" date. Here's claude's own summary:

R/InfinitySparseMatrix.R — dbind() function:

• Captures names(mats) before processing
• Builds a parallel group_labels vector during the conversion lapply: BISMs
  contribute their @groups level names, named entries use their list names,
  unnamed entries get NA
• Flattens group_labels alongside mats
• Replaces NA labels with position-based indices, incrementing to avoid
  collisions with existing named labels
• Uses factor(..., levels = unique(group_labels)) to preserve input order

tests/testthat/test.InfinitySparseMatrix.R — new "dbind uses meaningful group
names" test block:

• Named list → levels match list names
• Unnamed list → levels are "1", "2", ... (backward compat)
• BISM + named ISM → BISM keeps its group levels, ISM gets list name
• Partially named → named entries use names, unnamed fall back to position
  index
• Collision avoidance → unnamed entry skips indices already used by BISM
  groups
• Named list as single argument → inner names used

I'm going to read through the code and the test case. Then I'll do a linked PR.