Use Record's public API in a bunch of places (#10927)

# Description
Since #10841 the goal is to remove the implementation details of
`Record` outside of core operations.

To this end use Record iterators and map-like accessors in a bunch of
places. In this PR I try to collect the boring cases where I don't
expect any dramatic performance impacts or don't have doubts about the
correctness afterwards

- Use checked record construction in `nu_plugin_example`
- Use `Record::into_iter` in `columns`
- Use `Record` iterators in `headers` cmd
- Use explicit record iterators in `split-by`
- Use `Record::into_iter` in variable completions
- Use `Record::values` iterator in `into sqlite`
- Use `Record::iter_mut` for-loop in `default`
- Change `nu_engine::nonexistent_column` to use iterator
- Use `Record::columns` iter in `nu-cmd-base`
- Use `Record::get_index` in `nu-command/network/http`
- Use `Record.insert()` in `merge`
- Refactor `move` to use encapsulated record API
- Use `Record.insert()` in `explore`
- Use proper `Record` API in `explore`
- Remove defensiveness around record in `explore`
- Use encapsulated record API in more `nu-command`s

# User-Facing Changes
None intentional

# Tests + Formatting
(-)

crates/nu-cli/src/completions/variable_completions.rs

@@ -237,9 +237,9 @@ fn nested_suggestions(
     match value {
         Value::Record { val, .. } => {
             // Add all the columns as completion
-            for item in val.cols {
+            for (col, _) in val.into_iter() {
                 output.push(Suggestion {
-                    value: item,
+                    value: col,
                     description: None,
                     extra: None,
                     span: current_span,