Table spans

[teunbrand]

This PR stacks on top of #551.

The user-facing part of this is that it introduces the SPAN clause which can be used to create spanners.
This deliberately deviates from the original spec due to objections mentioned here: rich-iannone/gtsql#1.

TABULATE FROM ggsql:penguins
  SPAN 'Bill' ACROSS bill_dep, bill_len

Spanners appear simple on the outside but actually have a bunch of associated logic:

• It reorders columns to gather spanned columns together when possible
• When multiple spanners are declared, they each need to have levels so they do not collide
• We need to account for fragmented (disjoint) spanners that have >1 start and end column (when column order doesn't allow for contiguous spanners)
• The column labels need to be stretched into empty spanner territory.
• There is some bookkeeping around spanner IDs

Spanners span multiple columns, so table cells now can have >1 width which the html writer encodes in the colspan attribute. Because column labels can be stretched, table cells can have >1 height, which the html writer encodes in the rowspan attribute. Because the html writer renders the cells row-by-row, there is some bookkeeping going on to keep track of row-spanned cells. Together, I think this completes the layout logic of the writer.

[teunbrand]

Re: label stretching

Behaviour deviates in this PR from gt's behaviour, because I understand why column label cells are stretched, but not why they are only stretched for 1 extra row, and not all empty spanner area (why column d doesn't reach up to spanner Z for example).

library(gt)

df <- data.frame(x = 0, a = 1, b = 2, c = 3, d = 4, e = 5)
tb <- df |> gt() |> 
  tab_spanner("X", c(a, b)) |> 
  tab_spanner("Y", c(b, c)) |>
  tab_spanner("Z", c(c, d)) |>
  tab_style(cell_borders(), locations = cells_column_labels())
tb

^{Created on 2026-09-17 with reprex v2.1.1}

[teunbrand]

Another idea I had, but not implemented here is maybe we could do this:

TABULATE FROM ggsql:penguins
  SPAN 'Bill' ACROSS bill_len AS 'Length', bill_dep AS 'Depth'

So we could rename the spanned columns. Currently we have no grammar precedent for using identifier - AS keyword - string anywhere, so I didn't include this.