fix(deploy): bound captured command output

[andrewkcchung]

Why

Deployment and rendering tools can legitimately emit more output than Node's default synchronous-process buffer, causing otherwise successful operations to fail with ENOBUFS. A named, finite platform bound makes these commands reliable while keeping memory use explicit and preventing callers from selecting an unbounded capture size.

What changed

• Define a named 64 MiB maximum for each captured output stream.
• Apply the bound to the shared deployment command runner.
• Reject invalid or unbounded caller overrides.
• Add a regression that produces output above Node's default buffer and proves the shared runner captures it successfully.

Behavioral proof

Command:

node --test deploy/scripts/test/common.test.mjs

• Red — test-only patch on main: The large-output regression failed with ENOBUFS under the previous default-buffer behavior.
• Green — this branch: 21 tests passed, including the large-output and unbounded-value regressions.

Risk assessment

Relative risk: Low

The change adjusts one shared deployment command-runner resource bound without changing command construction or execution semantics. Memory remains explicitly capped at 64 MiB per stream. Tests reproduce the prior ENOBUFS failure and reject invalid or unbounded overrides; peak memory when both streams fill is the main review concern.

Review focus

• Confirm 64 MiB per stream is a reasonable finite ceiling for deployment render/build output.
• Consider peak memory when both stdout and stderr approach the limit concurrently.
• Confirm invalid overrides fail explicitly instead of silently reverting to an unsafe or unbounded value.
• Confirm existing argument redaction and environment-file parsing behavior remains unchanged.