fix(coderd): reject overflowing chat workspace TTL values - #31036

PUT /api/v2/chats/config/workspace-ttl converted workspace_ttl_ms to a time.Duration before it checked the 30-day maximum. A value whose nanosecond count overflows int64 wraps into range: 18446744133710 ms becomes about one minute, passes both range checks, and is stored. This PR checks the maximum on the millisecond value, before the conversion.

Why it is safe: for every value that does not overflow, req.WorkspaceTTLMillis > ttlMaximum.Milliseconds() is the same test as d > ttlMaximum, and it writes the same 400 message. TestChatWorkspaceTTL already pins the boundary: exactly 30 days is accepted and 721 hours is rejected. A mutant that rejects exactly 30 days fails it.

Red test: TestChatWorkspaceTTL now sends 18446744133710 ms and expects a 400 and an unchanged stored value. On main the handler accepts the value:

coderd: PUT path=/api/v2/chats/config/workspace-ttl status_code=204
exp_chats_test.go:17027: Error: An error is expected but got nil.
--- FAIL: TestChatWorkspaceTTL

Review focus: the order of the checks. The maximum now runs on milliseconds before the conversion; the one-minute minimum still runs on the converted duration, which is safe once the maximum has passed.

#31018 moves this handler into a shared putChatOperationalSetting helper. This fix is based on main, and whichever of the two merges second will merge main and keep this check.

Validation: go test ./coderd -run '^TestChatWorkspaceTTL$' -count=1 fails on main as above and passes on this branch. On this branch, go test ./coderd -run '^(TestChatPersonalModelOverridesAdminSettings|TestChatComputerUseProvider|TestChatDebugLoggingSettings|TestChatWorkspaceTTL|TestChatRetentionDays.*|TestChatDebugRetentionDays|TestChatAutoArchiveDays|TestAuditedChatOperationalSettingWrite.*)$' -count=1 passes (11 tests, 27 with subtests) and go test ./enterprise/coderd -run '^TestChatOperationalSettingsAudit$' -count=1 passes (17 with subtests). go vet ./coderd/ is clean, and the pre-commit hook passed (make gen, formatting, golangci-lint, slim build).

Xum, an AI agent, opened this PR for @ibetitsmike and wrote this description.