fix(telegram): bound fallback connection pools

[fmercurio]

Summary

• cap Telegram fallback IPv4 targets so lazy fallback-pool fan-out is finite
• distribute the configured connection and keepalive budget across both PTB clients and every primary/fallback pool
• make direct transport defaults transport-wide rather than per-IP pool budgets
• preserve explicit caller limits and existing fallback cleanup/CLOSE_WAIT protections

Validation

• focused Telegram network, fallback-pool-release, and CLOSE_WAIT tests: 36 passed
• compileall on touched Telegram modules: passed
• git diff --check: passed
• independent review: no blockers (base 4323c67; reviewed patch SHA-256 ad521d60e3ee5db978dc8148eeae7ffb5c2f8abe7ca5931c06b2bee1f9245884)

[alt-glitch]

This was generated by AI during triage.

Duplicate of #82860 — the live diffs implement the same fallback-pool cap and per-pool budget derivation.