fix: guard float() env var casts against ValueError (16 locations)

[annguyenNous]

Summary

Guards 16 unguarded float(os.getenv(...)) calls across 8 files against ValueError when env vars contain non-numeric strings.

Bug Pattern

# BEFORE (bug): crashes with ValueError if env var is "abc"
timeout = float(os.getenv("HERMES_API_TIMEOUT", "1800"))

# AFTER (safe): falls back to default on invalid input
timeout = _safe_float_env("HERMES_API_TIMEOUT", 1800.0)

The or default pattern (float(os.getenv("X") or 15)) does NOT protect against this — "abc" is truthy, so or never fires.

Files Fixed

File	Locations	Method
agent/chat_completion_helpers.py	3	_safe_float_env helper
agent/auxiliary_client.py	2	_safe_float_env helper
run_agent.py	2	_safe_float_env helper
hermes_cli/auth.py	2	_safe_float_env helper
hermes_cli/runtime_provider.py	2	_safe_float_env helper
gateway/platforms/telegram.py	1	inline try/except
gateway/platforms/wecom.py	2	inline try/except
plugins/platforms/discord/adapter.py	2	inline try/except

Impact

A typo in any of these env vars (e.g., HERMES_API_TIMEOUT=abc) currently crashes the entire agent/gateway on startup. With this fix, it logs a warning and uses the default value.

Test Plan

• All 8 modified files pass py_compile
• Set HERMES_API_TIMEOUT=abc and verify agent starts with default timeout
• Set HERMES_DISCORD_TEXT_BATCH_DELAY_SECONDS=xyz and verify Discord adapter starts

[teknium1]

Thanks for identifying a real malformed-environment failure. Most of this PR's original sites were independently migrated to the canonical helper in a7dd98c860; two runtime-provider paths still need a focused salvage.

Problems

• hermes_cli/runtime_provider.py:1441 and :1765 still call float(_getenv("HERMES_NOUS_TIMEOUT_SECONDS", "15")), so a non-numeric scoped value raises ValueError.
• The proposed _safe_float_env reads os.getenv, but current runtime_provider._getenv at hermes_cli/runtime_provider.py:49-59 is profile-scoped through get_secret. A salvage must retain that lookup path.
• run_agent.py:1257-1259 is a current sibling unguarded float conversion that should be included or tracked separately.

Suggested changes

• Retain only the two remaining runtime-provider fixes, parse the _getenv result safely, and add regression tests for both Nous resolution branches.

Automated hermes-sweeper review.