fix(ingest): correct --write-mode to the real load-mode vocabulary

[eddietejeda]

Closes #278.

Problem

ingest create --write-mode hardcoded value_parser = ["replace", "upsert"], which contradicted the rest of the CLI and the service:

• ingest sources fields <family> renders the server-reported per-family write_modes (e.g. the filesystem family: replace, append). So one CLI surface advertised append while ingest create --write-mode append failed with invalid value 'append' [possible values: replace, upsert].
• Confirmed against the live OpenAPI spec and the dlt worker: the real service-wide load-mode vocabulary is replace/append (keyless) plus the key-based upsert/update/delete (the worker maps upsert → dlt {disposition: merge, strategy: upsert}). The old pair both omitted valid modes (append, update, delete) and offered upsert as if universal, though it is key-gated.

Fix

• Correct the enum to the full service-wide set [replace, append, upsert, update, delete]. clap still catches typos up front, and every mode a family may advertise is now accepted; per-family/per-table support (e.g. a key-based mode on a keyless table) is enforced by the service.
• Rewrite the flag help to describe keyless vs key-based modes and point at ingest sources fields as the authoritative per-family source.
• Fix skills/hotdata/SKILL.md: the write_mode note listed only replace | upsert and wrongly said upsert needs a continuous bucket ingest — it needs a declared key on the destination table.

Before / after

# before:  hotdata ingest create ... --write-mode append
error: invalid value 'append' for '--write-mode <WRITE_MODE>'  [possible values: replace, upsert]

# after:   --write-mode append  -> forwarded to the service (proceeds)
# after:   --write-mode replce  -> invalid value 'replce'  [possible values: replace, append, upsert, update, delete]

Testing

• cargo build clean; cargo test --bin hotdata ingest → 82 passed, 0 failed.
• Verified on the built binary: append now parses and reaches the service; typos are still caught locally against the full enum.

cc @anoop-narang

[claude]

super nit: this doc comment is clap long-help, so hotdata ingest create --help will print "the flag forwards any value and lets the service validate, rather than hardcoding a list that drifts... see issue #278" to end users. The other issue-number references in this repo (indexes.rs:91, sdk.rs:1717, query.rs:314) are all in internal function docs and // comments, not in user-facing help strings. Consider keeping the user-relevant half and dropping the implementation rationale + tracker ref (or moving them to a // comment above the field, which is where the same argument in --record-shape effectively lives). (not blocking)

[claude]

nit: nothing pins the new behavior. The existing write_mode tests (destination_flags_default_schema_and_write_mode, ~line 2396) call build_destination directly and never go through clap, so re-adding a value_parser here would not fail any test — the exact regression #278 describes. src/commands/results.rs:165-173 already has the pattern for a clap-level parse test (Wrapper + try_parse_from); a one-liner asserting ingest create ... --write-mode append parses would guard it. (not blocking)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[claude]

Correctness of the change looks fine: the enum now covers the modes the server-reported write_modes can advertise (src/client/ingest.rs:1136,1197,1234 show replace/append), build_destination still defaults to replace (src/commands/ingest.rs:938), and the --destination JSON path is unaffected. SKILL.md text matches.

The two prior non-blocking threads (issue ref in the clap long-help, no clap-level parse test pinning --write-mode append) are still open; both remain non-blocking.

[claude]

Reviewed the full diff (the branch was force-pushed, so no since-last-review diff was available). The corrected value_parser list matches the server-reported write_modes vocabulary used elsewhere in the CLI (src/client/ingest.rs:1136,1197,1222,1234), the replace default in build_destination (src/commands/ingest.rs:938) is unchanged, and the SKILL.md text is consistent with the new flag help. No blocking issues.

The two nits from the prior cycle (issue-number reference inside the user-facing clap long-help, and no clap-level test pinning the accepted values) are still open but remain non-blocking.

[eddietejeda]

Live parse verification

• --write-mode append now clears clap and reaches the service (fails with datasource_not_found for a fake datasource id — i.e., the request was actually sent; before this PR it died locally with invalid value 'append').
• --write-mode replce → local clap error listing the full vocabulary [replace, append, upsert, update, delete].
• update and delete accepted by clap and forwarded.

No conflicts with main (GitHub reports CLEAN); CI green. LGTM.