Unify remaining fill paths under native settlement

[luisleo526]

Remaining partial, bound, reversal and RAW fill paths now use one native settlement owner. Source adapters resolve quantity, ordering and lifecycle policy before settlement; inspection, account projection and execution share allocation and fee logic.

• Add transient opening-identity selection and account projection without changing existing native request/event layouts or C ABI 4.
• Preserve exact reversal targets: closing a held position of 1 and opening 0.1 keeps the original 0.1 bits. Existing transaction-flow semantics remain unchanged.
• Translate source FIFO endpoints into complete physical opening selections, preventing unintended extra close rows while preserving exact native partial reductions and untouched sibling lots.
• Preserve historical entry costs, one current ticket, source slot policy and fill metadata; reject invalid accounting before mutation.
• Run local and CI verification through scripts/ci_verify.py: full builds, source guards, real historical ABI providers, CTest, installation and package smoke. Version identity, compiler/curl caches, cancellation and retained diagnostics are explicit and consistent.

Validation: all nine GitHub checks passed on the first attempt at 824d5d3. Release, Debug and sanitizer profiles each passed 329 tests on the applicable macOS/Linux lanes; native passed 337 with no skips, including required WebSocket coverage. Sanitizers ran with leak detection. Every profile passed all 22 real-library ABI link checks; historical e60 and 0e providers and the synthetic PR merge checkout were verified against the candidate tree. Independent review attests the exact final commit.

The fixed Cloud sweep completed 72 cases and all 4,190 observations/verifier reports with unchanged configuration, coverage and full grades. Four probe trade outputs differ from R2: XLM/BTC retain native reduction rows that improve matched coverage from 99% to 99.2%, and two Caldera probes have seven last-digit PnL/excursion changes reproduced by actual old/new libraries retaining exact remaining quantity and historical paid costs. The Caldera probes' other verifier metrics and cumulative PnL are unchanged. Root accepted these bounded contract consequences; slice-size parity on the two grids remains later campaign work.

The actual Cloud gate returned FAIL target.not-positive. Baseline/candidate snapshot grade objects are identical, and the active baseline pointer remained stable. The failure is retained under the user's standing non-regression refactor authorization; no PASS was fabricated and baseline promotion remains deferred. Merge uses squash and requires exact merged-tree equality plus post-merge CI. This closes physical settlement migration; remaining R3 observation/admission/clock ownership and R4 codegen lowering are separate roadmap work.