QVAC-23752 feat[bc]: drop n_discarded from the SDK config schema

[yingying0906]

🎯 What problem does this PR solve?

• The llm-llamacpp addon stops consuming n_discarded in QVAC-23752 feat[bc]: remove sliding-context support from the llm-addon #3938, so the key reaches llama's own argument parser and fails model load as an unknown option. Every in-repo consumer still sends it.
• parseContextOverflowMessage matched one of the addon's five numeric overflow wordings. The two that report a warm cache parsed as nothing, and on a warm cache it reported the appended prompt alone, which renders as "31 prompt tokens exceeds the 8192-token context window".
• isAddonContextOverflowError did not know about at batch prefill step, which both text guards and the MTMD batch guard emit, so on the message-only path those errors reached the client raw and instanceof ContextOverflowError failed.

📝 How does it solve it?

• n_discarded removed from both zod schemas, so it is no longer accepted or forwarded, and contract/schema.json plus the sdk-python client regenerated from them with the pinned generators. One field each.
• Removed from every e2e config that passed n_discarded: 256.
• toolBlockEvictable is gone. It existed because the addon's discard window opened exactly where the static tool block sat, so while sliding was possible the block had to travel with every turn. Nothing evicts it now, so a warm turn skips it whenever the prefix is known to hold a rendered one.
• Every overflow pattern now captures the context size last and the parts of the failing requirement before it, so promptTokens is the sum of the leading groups. One rule covers both shapes: the cached guards have two leading groups, the rest have one and the sum is that figure. Where a guard reports positions and KV cells, the cells are captured, since M-RoPE media occupies more cells than positions.
• The KV-cache guide told readers to add n_discarded for long conversations and named it as the fix for a prefill overflow. It now describes the two ways a caller runs out of context.

Ordering. This does not depend on #3938 shipping first. The old patterns are kept alongside the new ones, so the parsers match what 0.45.0 emits today and what #3938 will emit later, and no addon version pin moves here. Worth knowing for review: the e2e configs passed n_discarded: 256, which with the currently published addon turns sliding on, so dropping it means those runs no longer slide. That is the permanent behaviour once #3938 is released.

🧪 How was it tested?

• Context-overflow parser suite: 10/10 in both sdk and inference, with one case per addon guard using the exact emitted strings, the pre-existing older wordings kept as regression cases, and a case asserting the reported figure exceeds the window, which is what makes the formatted message coherent.
• KV-cache tool-block suite: 5/5 in both packages. The case that asserted a resend under sliding now asserts that a config still carrying the retired key does not force one, so the removal keeps a regression guard.
• contract:check and the sdk-python generate.py --check both clean, so the generated files match their sources.
• prettier and lint clean in both packages.

💥 Breaking Changes

n_discarded is no longer accepted in the llamacpp model config, and it is gone from the exported contract and the generated Python client.

BEFORE:

await loadModel({
  modelSrc: MODEL,
  modelType: 'llm',
  modelConfig: { ctx_size: 2048, n_discarded: 256 }
})

AFTER:

await loadModel({
  modelSrc: MODEL,
  modelType: 'llm',
  modelConfig: { ctx_size: 2048 }
})

[github-actions]

Review Status

Current Status: ❌ PENDING
Approvals so far: none

Pending reviews: Needs 1 Management or Team Lead, and 1 more from Management, Team Lead, or Member.

[github-actions]

License compliance — clean

No new dependency license findings in this PR.

Warn-only (shadow) mode — this check does not block merges yet.

Updated automatically by the canonical license compliance workflow.

NOTICE presence (advisory)

Missing NOTICE (advisory, does not block):

• ./.github/actions/release-merge-guard
• ./docs/website
• ./packages/ggml-coload-smoke
• ./packages/fabric/test/integration
• ./packages/inference-addon-cpp/mobile
• ./packages/sdk/e2e
• ./packages/llm-llamacpp/benchmarks/performance
• ./packages/llm-llamacpp/benchmarks/server
• ./packages/vla-ggml/sim/server
• ./packages/embed-llamacpp/benchmarks/performance
• ./packages/embed-llamacpp/benchmarks/server
• ./packages/asr-ggml/benchmarks/server