Exercise 2: A Genuinely Empty Text Field Round-Trips Correctly — Possible Solution
====================================================================

THE TEST
------------------------------
    schema4 = Schema([('note', 'TEXT')])
    row4 = [""]

    encoded4 = encode_row(schema4, row4)
    decoded4 = decode_row(schema4, encoded4)

RESULT
------------------------------
    encoded4 == b'\x00\x00'   (exactly 2 bytes)
    decoded4 == [""]           -- exact match

HOW MANY BYTES AN EMPTY STRING ACTUALLY PRODUCES
------------------------------
encode_text("") computes:

    encoded = "".encode('utf-8')          # b''  -- zero bytes, correctly
    length_prefix = struct.pack('>H', len(encoded))   # struct.pack('>H', 0) -> b'\x00\x00'
    return length_prefix + encoded         # b'\x00\x00' + b'' = b'\x00\x00'

The final record is exactly 2 bytes long: the length prefix itself
(storing the value 0), with zero content bytes following it, since
there's genuinely nothing to store for the text itself.

WHY THIS IS A VALID RECORD, NOT A SPECIAL CASE
------------------------------
Nothing in encode_text or decode_text branches on whether the string
happens to be empty -- the same code path that handles a 5-character
string handles a 0-character one, simply because len(b'') is a
perfectly ordinary, valid value (0) for the '>H' format to pack, and
data[offset:offset+0] is a perfectly ordinary, valid Python slice that
returns an empty bytes object. decode_text reads the length (0),
reads exactly 0 bytes starting at the current offset (getting b''
back), and decodes that empty byte string to "" -- again, no special
casing anywhere; Python's own b''.decode('utf-8') is simply defined to
return "".

WHY THIS WORKS AS AN ANSWER
------------------------------
This confirms the length-prefix design handles zero-length content as
a natural consequence of its own general mechanism, not as something
that needed to be anticipated and coded around separately. A real
database has to store empty strings constantly (an optional "notes"
field a user simply left blank, for instance) -- a record format that
required special-casing "what if this column has no content at all"
would be a design with a real gap in it; this one doesn't have that
gap, because the length prefix already generalizes correctly down to
zero.