Series build: a partial refusal exits 2, so a wrapper can still deliver

[mspinola]

The failure this fixes

TradingView restated two put/call closes. The build refused those two symbols exactly as designed, and because the only signal it gave was "non-zero", run-series.cmd exited before its replica syncs. Thirteen breadth series sat correct on the producer box and five sessions stale on both replicas until someone went looking.

A refusal has to stop the symbol, not the delivery of every other symbol.

The change

The build now says which kind of non-zero it is:

code	meaning	wrapper should
0	nothing refused	sync
2 (EXIT_PARTIAL)	some refused, others did not	sync, and surface the refusal
1	nothing usable: every target refused, a defer, a hard error	stop

The result dict carries partial beside ok, and the CLI prints a line telling a wrapper to run its syncs anyway.

A symbol that was already current counts as succeeding. The flag answers "did every symbol refuse", not "did this run write rows". Keying it on rows written would have left the live case blocked, because the thirteen good series were already current by the time the two bad ones began refusing. There is a test for exactly that.

Test plan

• Four new tests: partial exits 2 and prints the instruction; every-symbol-refusing stays 1; an already-current symbol still counts as succeeding; the flag is on the result dict. The refusal they trigger is the real one, a raw bar disagreeing with a stored bar.
• Full suite 308 passed, 14 skipped. Ruff clean.
• One existing test updated: it pinned the empty-targets result dict exactly, which now carries partial.

Wrapper side: mspinola/cotdata#119.

🤖 Generated with Claude Code