fix(matrix,simplex): tolerate malformed HERMES_*_TEXT_BATCH_* delay env vars

[ly-wang19]

Summary

Follow-up to a7dd98c86 ("guard remaining malformed int/float env var casts with utils helpers").

That sweep routed the HERMES_<PLATFORM>_TEXT_BATCH_* delay casts through the safe utils.env_float helper for Feishu, WeCom, Discord, and Telegram — but missed the identical sibling sites in the Matrix and SimpleX adapters, which still do a raw float(os.getenv(...)):

# gateway/platforms/matrix.py (MatrixAdapter.__init__)
self._text_batch_delay_seconds = float(os.getenv("HERMES_MATRIX_TEXT_BATCH_DELAY_SECONDS", "0.6"))
self._text_batch_split_delay_seconds = float(os.getenv("HERMES_MATRIX_TEXT_BATCH_SPLIT_DELAY_SECONDS", "2.0"))
# plugins/platforms/simplex/adapter.py
self._text_batch_delay = float(os.getenv("HERMES_SIMPLEX_TEXT_BATCH_DELAY", "0.8"))

A non-numeric value — e.g. a realistic unit-suffix typo HERMES_MATRIX_TEXT_BATCH_DELAY_SECONDS=0.6s — makes float("0.6s") raise an uncaught ValueError inside MatrixAdapter.__init__. That propagates through GatewayRunner._create_adapter and the unguarded platform-init loop in GatewayRunner.start(), so the whole gateway fails to boot when Matrix is enabled.

This is an overlooked miss, not an intentional fail-fast:

• the same __init__ already guards MATRIX_ROOM_IDENTITY_TTL_SECONDS with try/except + fallback;
• the sweep's own commit message states only sites already guarded by try/except or helpers were left untouched;
• the four sibling text-batch adapters were all converted — Matrix/SimpleX are simply the ones the sweep didn't reach.

Fix

Route both adapters' casts through env_float (returns the default on ValueError/TypeError), exactly as the four siblings were converted. No behavior change for valid values; the empty-string case also degrades to the default instead of raising.

Tests

tests/gateway/test_matrix.py::TestMatrixTextBatchEnvParsing — constructs MatrixAdapter with a malformed HERMES_MATRIX_TEXT_BATCH_DELAY_SECONDS / ..._SPLIT_DELAY_SECONDS and asserts init no longer raises and falls back to the default; plus a valid-value case. The two malformed cases fail without the fix (ValueError out of __init__).

[alt-glitch]

Related: #48771 (feishu env-guard), #48757 (wecom env-guard) — same env_float/env_int guard primitive applied per-platform. Also #35790 / #47338 (broader cross-adapter env-cast hardening). This PR fills the Matrix + SimpleX gap the a7dd98c86 sweep missed; the carried scope is the genuine per-PR fix.

[ly-wang19]

Rebased onto latest main to resolve the conflict. The Matrix adapter moved from gateway/platforms/matrix.py to plugins/platforms/matrix/adapter.py, so the env_float fix is now applied there, and the new TestMatrixTextBatchEnvParsing import was updated to plugins.platforms.matrix.adapter. The three tests pass locally.

[teknium1]

Thanks for carrying the missed Matrix and SimpleX env-float hardening forward. The underlying direct-construction defect is present on current main at plugins/platforms/matrix/adapter.py:926-931 and plugins/platforms/simplex/adapter.py:194-196; utils.env_float provides the intended fallback at utils.py:421-429.

Problems

• The added Matrix comments and test docstring say malformed input aborts gateway boot. Current plugin creation catches factory exceptions in gateway/platform_registry.py:318-328, and GatewayRunner.start() continues when no adapter is returned at gateway/run.py:7047-7060. Please describe the current consequence as Matrix adapter creation failing instead.
• The SimpleX production change has no corresponding malformed-env regression test. Its current adapter-init coverage lives in tests/gateway/test_simplex_plugin.py:121-142.

Suggested changes

• Correct the stale gateway-boot explanation in the added Matrix comments/tests.
• Add a SimpleX malformed-delay fallback test alongside its adapter-init tests.

Automated hermes-sweeper review.

[teknium1]

Current main no longer lets this exception abort the full gateway boot: PlatformRegistry.create_adapter() catches factory exceptions at gateway/platform_registry.py:318-328, and GatewayRunner.start() continues when it receives no adapter. Please describe this as failed Matrix adapter creation rather than a gateway-wide crash.

[teknium1]

Please add a malformed HERMES_SIMPLEX_TEXT_BATCH_DELAY regression case in tests/gateway/test_simplex_plugin.py alongside the existing adapter-init tests, so this changed fallback path is covered as well as Matrix.

[ly-wang19]

Rebased onto current main. The Matrix adapter has since moved (gateway/platforms/matrix.py → plugins/platforms/matrix/adapter.py), so this is re-targeted at the new location — the raw cast is still there today:

plugins/platforms/matrix/adapter.py:950   os.getenv("HERMES_MATRIX_TEXT_BATCH_DELAY_SECONDS", "0.6")
plugins/platforms/simplex/adapter.py:195  os.getenv("HERMES_SIMPLEX_TEXT_BATCH_DELAY", "0.8")

Both now go through utils.env_float, matching how Feishu/WeCom/Discord/Telegram read the same HERMES_<PLATFORM>_TEXT_BATCH_* family after a7dd98c. Matrix test suite is green (the one unrelated TestMatrixProxyConfig::test_no_proxy_by_default failure reproduces on unmodified main in any environment with HTTP_PROXY set, so it's not from this change).