Fix integer limit in PDU definitions

[jplatte]

Resolves #2485.

[turt2live]

@jplatte can we get sign-off for this change please? Just need to leave a comment like the following:

Signed-off-by: Your Name <email@example.org>

[jplatte]

Signed-off-by: Jonas Platte <jplatte+git@posteo.de>

[richvdh]

Having implied this is the right solution in #2485, I'm not sure it's as simple as all this. There are two problems:

• There certainly exist rooms with depths greater than 2**53, and if we suddenly change the rules, they'll stop working (or at least they would, if we honoured the spec).
• In general, having the spec differ from what synapse does in practice is unhelpful: it just means that other implementations either have to ignore the spec, or fail to interoperate with synapse.

The real solution here is to not have to rely on the validity of depth (see MSC1229); in the shorter term I'm not entirely sure. We could either leave it as-is for current room versions and change it for future room versions, or we could take the hit on breaking the rooms with depths greater than 2**53 (which will only be that way due to past malicious abuse) and change the implementation in synapse.

I think that changing the spec without making synapse match is the worst of all possible worlds.

[turt2live]

We are fixing this in a different way, represented by #2540