fix(postgrest): normalize boolean values in is_() to lowercase literals

[winklemad]

No linked issue — found by reading the postgrest filter builder.

What kind of change does this PR introduce?

Bug fix.

What is the current behavior?

is_() normalizes Python None to the PostgREST literal null, but does nothing for Python booleans. A boolean therefore falls through to filter(), which formats it into the query string as is.True / is.False:

client.table("t").select("*").is_("active", True)
# -> ?active=is.True

PostgREST's is operator is case-sensitive and only accepts the lowercase trilean literals null, true, false, unknown (see PostgREST/postgrest#2077). is.True is rejected by the server, so .is_(col, True) / .is_(col, False) — the natural way to filter a boolean column against IS TRUE / IS FALSE — produce a failing request.

What is the new behavior?

is_() converts True/False to "true"/"false", using is True / is False identity checks so only real booleans are touched — strings such as "unknown" / "not_null" and any other value pass through unchanged. This matches supabase-js, whose .is('col', true) emits is.true.

client.table("t").select("*").is_("active", True)   # -> ?active=is.true
client.table("t").select("*").is_("active", False)  # -> ?active=is.false
client.table("t").select("*").is_("active", None)   # -> ?active=is.null   (unchanged)

Additional context

The fix lives in the shared base_request_builder.py, so it covers both the sync and async clients. Added test_is_true / test_is_false / test_is_none to the postgrest filter-builder tests (_async and _sync). The full postgrest unit suite passes and ruff check + format are clean.