fix(alibaba-token-plan): correct deepseek-v4.1-flash overrides for the plan gateway

[oliver-mee]

What changed

Modifies providers/alibaba-token-plan/models/deepseek-v4.1-flash.toml (added via #7147) to correct
two fields against measurements taken on the live Token Plan gateway. This is now an override-only
diff against the merged row, matching how the file landed.

1. structured_output = false. The lab row (models/deepseek/deepseek-v4.1-flash.toml) carries
structured_output = true, and provider rows inherit omitted fields. On the international Token
Plan gateway, however, response_format: json_schema is rejected for this model id with HTTP 400
"This response_format type is unavailable now" in both thinking modes, while json_object works.
Probed 2026-09-15 on both Personal and Team editions (Singapore). Without the explicit override the
provider row advertises a capability the plan gateway refuses.

2. Reasoning-effort contract note. The header comment now records what the plan gateway
actually accepts: the string ladder low / high (default) / max, with medium/xhigh accepted
as aliases, and the fact that the integer effort values the vendor OpenAI-chat reference publishes
are rejected by this gateway for this id (probed 2026-09-15). Same values as #7147, richer sourcing.

Evidence

Claim	Source
json_schema 400 on plan gateway, both thinking modes	Gateway probe 2026-09-15, Personal + Team editions, error text quoted in the file comment
json_object accepted	Same probe session
effort ladder low/high(max default), integers rejected	Same probe session; enum observed from the gateway's own 400 message
Lab row inheritance semantics	Upstream merge behaviour: an omitted provider field takes the lab value

Related: #7146 (the issue that prompted #7147).

---

🤖 Generated with Hermes Agent (Qwen3.8-Max, Alibaba Token Plan)
Co-authored-by: Hermes Agent

[github-actions]

Action items

• [medium] [violation] providers/alibaba-token-plan/models/deepseek-v4.1-flash.toml:16 - Check: Provider overrides must record real capability deltas after base_model merge. Why: Lab deepseek/deepseek-v4.1-flash has structured_output = true. Omitting the field inherits that value, so the resolved Token Plan model advertises structured output even though the PR body and leading comments state this host rejects response_format json_schema (400). Omission does not clear the capability. Action: Add structured_output = false as a provider override (same pattern as other hosts that lack dedicated structured output for this model).

[github-actions]

No actionable findings.

[oliver-mee]

Superseded by #7151: my branch rebuild (to rebase onto current dev after #7147 merged) made this PR's head commit unreachable, which closed it irreversibly. The replacement PR carries the identical single-file change against the post-#7147 state.