fix: restore cursor to correct position on picker close

[jamestrew]

I think we have several factors contributing to flaky (off by one error) behaviors in the cursor position on picker close.

Mainly, I think we should be scheduling nvim_win_set_cursor. Without doing so, we've had to "arbitrarily" +1 on the columns to compensate:

telescope.nvim/lua/telescope/actions/set.lua

Lines 214 to 216 in 87e92ea

	col = pos[2] + 1
	elseif row == nil then
	row, col = pos[1], pos[2] + 1

This is made worse by a couple of other factors

1. ripgrep gives columns in 1-index whereas nvim_win_set_cursor expects 0-index.
   This wasn't previously an issue probably due to not scheduling nvim_win_set_cursor but I think we were just depending on UB.
2. not accounting for original mode on select_default. like this
   I think this is no longer necessary and was used to compensate for the lack of scheduling set cursor.
   

   telescope.nvim/lua/telescope/actions/init.lua

   Lines 388 to 390 in 87e92ea

   	if cursor_valid and a.nvim_get_mode().mode == "i" and picker._original_mode ~= "i" then
   	pcall(a.nvim_win_set_cursor, original_win_id, { original_cursor[1], original_cursor[2] + 1 })
   	end

I recognize this is a very sensitive part of our code that we've had repeat issues with so I still need to do some testing and verify this closes a few issues without any obvious or previously experienced regressions

closes #2319 and semi-related #2319 (comment)
closes #2849
closes #1366

TODO (all with both norma/insert mode as original mode):

• check 0/1 index of columns in lsp/treesitter/etc pickers
  • all except treesitter picker was using 1-index row/col (treesitter cursor position was actually broken) so taking 1-index as the true column index
• closing pickers without selecting anything (no change)
• select_default with no row/col info (find_files)
• select_default with row/col/both info (live_grep with/without --invert-match for no col info)
• actions.close() calls (like in Wrong cursor position after actions.close #2849)
• strange cursor move when telescope closes #2319
• anything else?

[Conni2461]

let me run this for a couple of days before merging :D but thanks for addressing this issue :)

[Conni2461]

i rebased the branch and continue to run it till friday, if there arent any issue till then, i think this is good to go thanks :)

[jamestrew]

I found an regression with :Telescope help_tags related to opening an existing buffer at a different location.

1. :Telescope help_tags -> search for bufhidden -> this opens the options.txt doc buffer
2. :Telescope help_tags -> search for hidden -> same options.txt but doesn't move, still on bufhidden

[jamestrew]

I found an regression with :Telescope help_tags related to opening an existing buffer at a different location.

1. `:Telescope help_tags` -> search for `bufhidden` -> this opens the `options.txt` doc buffer

2. `:Telescope help_tags` -> search for `hidden` -> same `options.txt` but doesn't move, still on `bufhidden`

The problem is that we're scheduling cursor position inside actions.close() so when a picker calls actions.close() then does something that moves the cursor under the current window, the previously scheduled cursor position moves the cursor again.

That why I had to schedule this so it happens after the cursor position is reset
https://github.com/nvim-telescope/telescope.nvim/pull/2850/files#diff-f192865bab9995c5a69e48b85b5d4534094c885a1556b10bb1e9407b0e0570d0R387-R390

For this help_tags issue, I can fix it by doing the same by scheduling this

telescope.nvim/lua/telescope/builtin/__internal.lua

Lines 778 to 784 in da8b3d4

	if cmd == "default" or cmd == "horizontal" then
	vim.cmd("help " .. selection.value)
	elseif cmd == "vertical" then
	vim.cmd("vert help " .. selection.value)
	elseif cmd == "tab" then
	vim.cmd("tab help " .. selection.value)
	end

Ideally, we just defer/schedule everything after actions.close() to happen after cursor position is set. I gotta think about how to do this without significant refactors...
I don't like the idea of throwing vim.schedule case by case on things after actions.close(). This really puts the responsibility on each picker actions way too much.

[jamestrew]

Yeah idno how to cleanly fix this without API changes and big refactors.