Firestore sink - #377

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@connectors/firestore/connector_integration_test.go`:
- Line 199: The test currently ignores the error returned from encodeJSON
(encoded, _ := encodeJSON(doc)), which can hide setup failures; update the test
to check the error and fail fast—e.g., replace the ignored error with error
handling using the test helper (t.Fatalf or require.NoError) to assert
encodeJSON(doc) returns no error and only then use the encoded value; reference
the encodeJSON call and ensure the test fails immediately if encoding fails.
- Line 180: The deferred type assertion connector.(interface{ Teardown()
}).Teardown() can panic if connector lacks that exact Teardown() method; change
it to perform a safe comma‑ok assertion inside the deferred function (e.g.,
defer func() { if td, ok := connector.(interface{ Teardown() }); ok {
td.Teardown() } }) so Teardown is only called when implemented and the test
won’t panic unexpectedly; optionally log or t.Log when Teardown is absent for
visibility.

In `@go.mod`:
- Line 45: Update the grpc dependency entry "google.golang.org/grpc v1.76.0" in
go.mod to v1.80.0 and ensure the module graph is refreshed (e.g., run go get
google.golang.org/grpc@v1.80.0 and go mod tidy) so the fix for CVE-2026-33186 is
applied; afterwards run the test suite/build to verify no regressions.

---

Nitpick comments:
In `@connectors/firestore/connector_test.go`:
- Around line 104-125: Update TestValueToString to include BSON and error
scenarios: add table entries that pass a bson.ObjectId (expecting its Hex()
string) and a bson.Binary (expecting hex-encoded data) to valueToString and
assert the returned strings, add a case with an unknown/complex type (e.g., a
custom struct) to assert the fallback fmt.Sprintf("%v") behavior, and add at
least one input that should cause valueToString to return an error and
assert.Error; reference the existing TestValueToString and valueToString
function names when locating where to add these cases.

In `@connectors/firestore/connector.go`:
- Around line 91-93: The code silently clamps invalid settings.BatchSize to
DefaultBatchSize; update the if block that checks settings.BatchSize to emit a
warning before changing the value so users know their configured value was
adjusted—e.g., log a message referencing the original settings.BatchSize and
DefaultBatchSize (use the connector's existing logger, e.g., logger.Warnf or
log.Printf) then set settings.BatchSize = DefaultBatchSize. Ensure you reference
settings.BatchSize and DefaultBatchSize in the warning so it's clear which
values were changed.
- Around line 322-325: The current switch cases for numeric types (case
int/int32/int64 and case uint/uint32/uint64) can cause precision loss for values
> JS_SAFE_INT (2^53-1); update the handling so uint64 (and unsigned types
promoted to uint64) are detected and for values > 9007199254740991 you return
their decimal representation as a string (use strconv.FormatUint(val, 10)),
otherwise return the numeric formatting; specifically modify the uint64 handling
branch (and any code paths that cast to uint64) to compare against the JS max
safe integer and return a string for large values to avoid downstream JSON
precision loss.
- Around line 199-220: In writeBatch, include context about which item failed by
changing the loop to capture the item index and include it in the returned error
when extractDocumentIDAndData fails; update the error returned from
extractDocumentIDAndData inside writeBatch (and any subsequent errors) to wrap
the original error with fmt.Errorf including the index (and optionally a short
snippet or hex of raw) and the collectionName so it's straightforward to
identify the problematic document when debugging; reference conn.writeBatch and
extractDocumentIDAndData to locate and modify the error wrapping logic.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@connectors/firestore/connector.go`:
- Around line 250-282: The switch currently treats UPDATE_TYPE_PARTIAL_UPDATE as
a full upsert; add an explicit case for
adiomv1.UpdateType_UPDATE_TYPE_PARTIAL_UPDATE in the switch and handle it by
converting update.GetData() via rawToMap (same as other branches), deleting
idKey, and calling the incremental update method on the batch writer (e.g.,
bw.Update(docRef, partialData)) with the same bw.End() and error wrapping
pattern; if the batch writer does not support partial updates, return a clear
error instead of falling through to bw.Set to avoid overwriting full documents.
- Around line 435-451: The valueToString function currently falls back to
fmt.Sprintf("%v", val) which silently converts complex types (maps, slices,
structs) into unstable IDs; change valueToString to only accept explicit types:
string, bson.ObjectID (Hex), integer/unsigned/float families (formatted),
bson.Binary (hex of Data), []byte (hex), and any type implementing fmt.Stringer
(use .String()); for any other type return a descriptive error like "unsupported
id type: %T" instead of the %v fallback so callers can handle invalid ID types
instead of producing unstable document IDs.