fix(bigtable): avoid self-referential error on RPC-level mutate failure

[kyungseopk1m]

When Table#mutate / Table#insert hits an RPC-level failure with entries still pending and no per-entry mutation errors, mutateInternal set err.errors to an array containing err itself. The resulting cycle (err.errors[i] === err) makes the error impossible to serialize with anything that follows the AggregateError convention of recursing into errors — for example pino, which throws RangeError: Maximum call stack size exceeded.

This pushes a fresh error carrying the RPC-level code and message for each pending entry instead of err itself. The errors array is only consumed for its per-entry code/message (the existing Table#mutate error tests and the callback.err.errors docs), so nothing relies on the former self-reference.

The existing "not retryable" test already exercises this branch (an index-0 success followed by a stream error); it now also asserts that err.errors[0] is not err and that the resulting error can be serialized.

• Make sure to open an issue as a bug before writing your code, and reference it in your PR.
• Ensure the tests and linter pass
• Code coverage does not decrease (if any source code was changed)
• Appropriate docs were updated (if necessary)

Fixes #8075 🦕

[gemini-code-assist]

Code Review

This pull request resolves a serialization issue in mutateInternal.ts where a self-referencing error was created by pushing the parent RPC error into its own errors array. It fixes this by creating a fresh entryError with the copied message and code, and adds a corresponding regression test. The reviewer suggested also copying metadata and details from the original RPC error to preserve debugging context.