* THANKS TO Yawning Angel for suggesting the clarification.

 * THANKS TO eikovi@sigaint.org for pointing them out.

Too complex and not sufficient gain. For full rationale, please see thread:
       https://lists.torproject.org/pipermail/tor-dev/2016-March/010560.html

- Kill last remnants of TAP from the proposal.

- Replace SHA256 with SHA3-256 and our KDF with SHAKE.

- Make the INTRO_ESTABLISHED cell extensible.

- Improve the descriptor format a bit.

- Don't be ambiguous about "INTRODUCE" cells (pointed out by malekbr).

- Cleanup the scaling section.

Signed-off-by: David Goulet <dgoulet@ev0ke.net>

Base64 holds 6 bits, not 5. Patch from twim.

Signed-off-by: David Goulet <dgoulet@ev0ke.net>

(Note that no corresponding tor change is needed, since we don't
actually use these keys yet.)

Resolves 17668.{S1,T1}

Signed-off-by: David Goulet <dgoulet@ev0ke.net>

After the following development meeting discussing this proposal:
http://meetbot.debian.net/tor-dev/2016/tor-dev.2016-02-04-13.28.html

These changes were sent to tor-dev@lists.torproject.org:
https://lists.torproject.org/pipermail/tor-dev/2016-February/010379.html

Signed-off-by: David Goulet <dgoulet@ev0ke.net>

- Remove ed25519 keys completely. Use RSA keys for referencing.
- Replace SHA256 with SHA3-256 (Keccak).
- Specify better the format and contents of TIMESTAMP.
- Put TIMESTAMP in the front of COMMIT for symmetry with REVEAL.
- Use base64 not base32.
- Specify what happens when PREVIOUS_SRV is unknown.
- Remove some paragraphs that are no longer valid.
- Simplify consistent ordering in HASHED_REVEALS.
- Put algorithm name first on disk and commit.

Update the directory spec to describe client behaviour with default
fallback directory mirrors after #15775 and #4483.

Signed-off-by: David Goulet <dgoulet@ev0ke.net>