id(web): prosemirror_to_markdown keeps only the first text node of a code_block — multi-node blocks are truncated on save #882

Bug

In pkgs/id/src/web/markdown.rs, json_to_ast's "code_block" arm reads only the first text child of the ProseMirror code block:

// markdown.rs:~638-642 (main 45c817e0)
let content = json["content"]
    .as_array()
    .and_then(|arr| arr.first())
    .and_then(|n| n["text"].as_str())
    .unwrap_or("");

A code block whose ProseMirror content is split into more than one text node — which editors produce legitimately, e.g. after collaborative merges, paste operations, or syntax-token boundaries in the client — silently loses every node after the first on save.

Verified repro (this run, main 45c817e)

Temporary probe test run with cargo test --features web --lib web::markdown:: (probe removed after the run; assertion below is the acceptance test):

ProseMirror input:

{
  "type": "doc",
  "content": [{
    "type": "code_block",
    "attrs": {"params": "rust"},
    "content": [
      {"type": "text", "text": "fn a() {}\n"},
      {"type": "text", "text": "fn b() {}"}
    ]
  }]
}

prosemirror_to_markdown emits:

```rust
fn a() {}
```

fn b() {} is deleted with no error. (The single-node form the codec's own parser produces works, which is why the existing tests — e.g. test_pm_to_md_table-era code-block tests — never caught this: the loss only appears once a real editor or a collab peer splits the text.)

Why it matters

This is the save path for the id web editor, and id is a collaborative system: a code block edited by multiple peers is exactly the content most likely to arrive as several text nodes. Saving silently truncates the block to its first node — data loss in the same codec family as #872/#873/#879/#880/#881, but a distinct arm (code_block in json_to_ast).

Suggested fix

Concatenate the text of all children of the code block (skipping/handling non-text children the same way the first-child code does today — i.e. treat a missing text as empty rather than aborting), e.g.:

let content = json["content"].as_array().map_or(String::new(), |arr| {
    arr.iter().filter_map(|n| n["text"].as_str()).collect()
});

and keep the existing format!("{content}\n") literal construction.

Acceptance criteria

• Regression test: a code_block with two or more text children round-trips with all text present, in order.
• Single-node and empty code blocks behave exactly as before.
• Existing web::markdown tests still pass (cargo test --features web --lib web::markdown::).
• Coordination: PR fix(id-web): img attribute codec — sequential parse, escaped emission (#872, #873) #878 (for id(web): parse_img_tag substring search lets data-src/data-width shadow the real img attributes — save rewrites the image URL #872/id(web): prosemirror_to_markdown emits img src/alt/title unescaped — quotes truncate values and can inject attributes on reparse #873) is in flight in the same file — sequence or rebase so the img-codec changes do not conflict.