fix(telegram): recover final delivery after stream flood

[kshitijk4poor]

Summary

Telegram final delivery now recovers after a flooded streaming edit without deleting earlier finalized stream segments or waiting through a second long flood cooldown.

Changes

• Preserve @helix4u's final-edit flood fallback and empty-tail final commit.
• Scope empty-tail cleanup to the active stream segment, preserving pre-tool preambles/commentary.
• Keep Telegram's structured long-flood delay, but leave delivery to the normal gateway path instead of sleeping through a long retry in the stream task.
• Add regression coverage for the real Telegram flood result, long-delay behavior, and tool-boundary cleanup.

Validation

Check	Result
scripts/run_tests.sh targeted streaming suite	164 passed
Real adapter long-flood contract + segment cleanup smoke	passed
Ruff / py_compile / git diff --check	passed
Codex final review	clean

Attribution

• fe7522fbb — @helix4u's original final-delivery recovery implementation, preserved by cherry-pick.
• 286040d63 — follow-up hardening for reviewer-found cleanup and flood-retry defects.

Closes #62253
Related: #55869, #59864, #54331

[alt-glitch]

This was generated by AI during triage.

Salvage / superseding of #62253 (@helix4u): this PR preserves #62253's original final-delivery recovery commit (fe7522fbb) via cherry-pick and adds follow-up hardening (286040d63) for reviewer-found cleanup/flood-retry defects. Same files and mechanism as #62253 (gateway/stream_consumer.py + Telegram adapter + final-delivery regression test), so this is related_to #62253, not a duplicate. Both are open; a maintainer should pick which to merge (this one is the superset with the extra hardening + segment-scoped cleanup). Same Telegram flood-control final-delivery family as #55869 / #59864 / #54331.