[ghcp-handoff] T10: Retry / resumability after timeout or transient failure

[bminier]

Problem

Copilot timed out at 10 min? Network hiccup mid-run? Today the user starts from scratch — the prompt file survives but nothing else.

Approach

`/ghcp-handoff retry <slug|last>`:

• Reuses the existing prompt + worktree
• Re-invokes Copilot with the same flags
• Preserves invocation history (append to `invocation_history: [...]` in .meta.json rather than overwriting)

Pairs with G3 (amend mode) — retry is amend with no modifications.

Review reference

Adversarial review T10

[bminier]

Addressed on branch ghcp-t10-retry-history (stacked on ghcp-t8-expand-triggers).

The retry mechanism itself shipped with #8 as ghcp-manage.ts rerun, but without history preservation. This commit closes the gap:

• ghcp-invoke.ts gains --preserve-history — when set, the prior invocation gets pushed onto invocation_history[] before the new one is written. Decoupled from --resume-session so retries (which use a fresh session) also keep lineage.
• rerun's suggested command now includes --preserve-history, so retrying after a timeout/crash keeps the failure record alongside the success.

Will close when merged to main.