Internal changes to tool execution: memory sampling and a waitForExitAfterKill option

What

Several internal changes to how Claude Code runs tools (the actions Claude takes, such as running a command or editing a file) and shell commands:

• Memory use around each tool call is now taken from a sampler, with sample("before_call"), sample("after_call") and sample("after_throw"). Before, it called process.memoryUsage() and computed rss, heap and external memory changes. The usage reports for tool success and error use the sampler's fields. Success reports also get fields describing the shape of the tool's result.
• The helper that runs processes has a new option, waitForExitAfterKill, which is off by default. The exec wrapper passes it through.
• The updater for background jobs still in progress gains an includeSettled option.
• The parser for artifact conflict records no longer rejects records for one extra condition.
• The bridge and CCR heartbeat gates are unchanged apart from renamed internals.

Why

These are mostly internal. Memory is no longer measured directly on every tool call. waitForExitAfterKill may help avoid stray output after a command that was stopped for taking too long.