fix code() failing with "line number not found in file stdio" on pasted input

[d-torrance]

readline 8+ enables bracketed paste mode by default. When the user pastes multiple lines at once, readline() can return a single string containing embedded newlines (e.g. "expr1\nexpr2") rather than one line per readline() call.

The old code called add_history() on the entire returned string, so the whole paste was stored as one history entry. Meanwhile M2's parser incremented lineNumber for every '\n' it saw, creating a mismatch: after a two-line paste the session line counter was 2 but only one new history entry existed.

code.m2 maps M2 line numbers to history entries via
getHistory(lineNumber + historyOffset)
so getHistory(2 + historyOffset) would call history_get() with an out-of-range index, return NULL, and ultimately produce:

error: line number 1 not found in file stdio

Fix: instead of a single add_history(p), split p at each embedded '\n' and add each segment as its own history entry (save/restore the '\n' in-place). For single-line input (the common case) this is identical to the previous behaviour.

Fixes: #4356

AI Disclosure

This was 100% Claude Code

[mahrud]

I don't think Claude identified the correct solution. We don't want to split user's input into several lines. It is a feature that if I paste multiple lines, then want to redo the same thing, I can go up until I get to the full multiline input and enter it again.

I'm not even sure Claude identified the correct problem. e.g. this works fine:

i1 : g = () -> (
    L = {};
    for i to 4000 do (
        L = append(L, i)
        )
    )

o1 = g

o1 : FunctionClosure

i2 : code g

o2 = stdio:1:4-6:5: --source code:
     g = () -> (
         L = {};
         for i to 4000 do (
             L = append(L, i)
             )
         )

Which indicates that lines 1 through 6 were correctly retrieved from history.

[pzinn]

I'm a little confused what this has to do with readline -- neither emacs nor WebApp mode use readline, and I've definitely encountered some version of this bug.

[d-torrance]

They do use readline a bit -- in particular for grabbing the code of things defined in stdio