fix(bitswap): blockpresencemanager leak

[guillaumemichel]

Summary

• Originated from bitswap: long-running session leaks #752
• Addresses BlockPresenceManager leak uncovered by @Wondertan in 7adce5f
• Modifies when to send out cancel (as introduced in Always send cancels even if peer has no interest #829)

Reasoning

• IIUC before Always send cancels even if peer has no interest #829, no cancel was sent directly after receiving a block.
• Maybe I misunderstood, but in Always send cancels even if peer has no interest #829, I don't see a reason why the cancels should be sent out before filtering the wanted keys. IIUC the filtering (in SessionsInterestManager) doesn't use the PeerManager wantlist.
• Cancels have to be sent out for blocks that we receive, but they can be sent out later.

@gammazero I suspect this PR is equivalent to #829 wrt. sending Cancels, although we don't send cancels for blocks we don't want. I suggest we run a short test to be convinced of it before merging.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 60.48%. Comparing base (9e257a7) to head (e34cfd8).
Report is 1 commits behind head on main.

@@            Coverage Diff             @@
##             main     #833      +/-   ##
==========================================
+ Coverage   60.44%   60.48%   +0.04%     
==========================================
  Files         244      244              
  Lines       31125    31120       -5     
==========================================
+ Hits        18812    18822      +10     
+ Misses      10630    10622       -8     
+ Partials     1683     1676       -7     

Files with missing lines	Coverage Δ
bitswap/client/internal/session/session.go	90.56% <100.00%> (+0.37%)	⬆️
...p/client/internal/sessionmanager/sessionmanager.go	95.95% <ø> (-0.20%)	⬇️

... and 9 files with indirect coverage changes