Make SizeLimit tests deterministic by deduplicating generator output

[dwcullop]

Problem

SizeLimitFixture.InvokeLimitSizeToWhenOverLimit and SizeLimitFixture.AddMoreThanLimitInBatched fail intermittently in CI with errors of the form:

Expected _results.Messages[1].Adds to be 10 because Should be 10 adds in the second update, but found 9 (difference of -1).

Both tests build a 10-item batch, then build a second 10-item batch, and assert that the two AddOrUpdate calls produced exactly 10 Adds each.

RandomPersonGenerator.Take(N) draws random Person rows from a finite name pool (~21 first names for girls + ~30 first names for boys, each cross-joined with 24 lastnames squared). Person.Key is Person.Name. Two independent .Take(10) calls can return rows that share the same generated Name, so the second batch's AddOrUpdate sees one of its rows as an existing key and emits 9 Adds + 1 Update instead of 10 Adds + 0 Updates.

The probability is small per row, but two Take(10) calls draw 20 rows into the same key space; birthday-paradox math puts the collision rate at roughly 0.3%. That matches the observed CI failure frequency for these two tests.

Fix

Draw a larger candidate pool (60 rows), dedupe by Key, take the first 20 distinct, and split into two non-overlapping batches of 10. No production code changes.

Verification

50 consecutive runs of SizeLimitFixture on the branch, zero failures. Targeted command:

for ($i=1; $i -le 50; $i++) {
  dotnet test ... --filter "FullyQualifiedName~SizeLimitFixture"
}